1. INTRODUCTION

Understandably, most people want less crime. Improvements to outdoor lighting are frequently mentioned in election promises, the news media and government planning documents as an option or action for crime reduction. Unfortunately, the common understanding of ‘improvements’ as ‘more and brighter’ in this context is likely to lead to ineffective or even counterproductive outcomes, as will be seen.

1.1 Rationale for this Study

Crime tends to be concentrated in cities, a fact that continues to attract the attention of criminologists (eg Glaeser and Sacerdote 1999, Ousey 2000)). Urban crime was already a problem before the widespread application of artificial lighting. This calls for an explanation as to why the present line of work has been pursued at all. The reasons are:

· It appears to be widely believed that more and brighter outdoor lighting would help to reduce both fear of crime and actual crime at night.

· Field experiments have been claimed as showing that increased outdoor artificial lighting does reduce actual crime.

· Publication of these claims reinforced existing beliefs, strengthening conflicts with separately established needs for environmental limits to outdoor lighting.

· In Part 1 of this work (Clark 2002b), study of the lighting and actual crime evidence indicated the claimed benefits were exaggerated or untenable.

· The facts uncovered went further than this, indicating that increased lighting and increased crime often occurred together.

· There are plausible mechanisms whereby the presence of light at night could facilitate some of the processes thought responsible for the growth of urban crime.

· Study of the nexus between outdoor lighting and crime now holds promise of containing the environmental as well as social problems of excessive outdoor light at night.

1.2 Results from Part 1

Part 1 of this work is a critical review of the literature on outdoor lighting and crime. It is summarised here as a foundation for this second part.

The literature indicates that artificial light at night may reduce the fear of crime to levels approaching those experienced in the same location in daylight. Illuminance of about 20 lux appears to be fully or highly effective in practice, and less is often adequate.[3] More light seems to be required in the presence of glare, or less without glare.

If there is some substance to the belief that more and brighter outdoor lighting would help reduce actual crime, then extending the belief to its ultimate stage means there should be little or no crime in the bright outdoor lighting conditions of daytime, but that is far from the facts. For example, 54% of violent crime in the USA occurred between 6 am and 6 pm, and only 20% of rapes involve unknown assailants at night (BJS 1999). Only 35% of all burglaries in the USA are reported to have occurred at night, or 48% of all burglaries for which the time of occurrence is known (UCR 1996). Note that these figures are for all reported crime in the whole of the USA, which gives them substantial face validity. In England and Wales, crime survey results indicated that nearly two-thirds of the total of assaults, burglaries, robberies, thefts and vandalism occurred after dark (Ramsay and Newton 1991, ONS 2000). An unresolved issue is the extent to which the proportion of crime at night is influenced by effects of light at night rather than by other factors such as the available amount of leisure time at night.

Effects of light on actual crime have been investigated with field studies, but extraneous influences have often confounded the results. Crime-reducing, nil, uncertain, and crime-increasing effects of light at night have variously been reported for night or day or both. Thorough scientific reviews published in 1977 and 1997 in the USA concluded that the effect of lighting on actual crime was unknown. Nevertheless, crime prevention specialists and some academics have asserted that lighting is important, at least in some circumstances, in the fight against crime. Recent UK findings of beneficial effects of brighter lighting have bolstered this view despite complaints, by others, of procedural and analytical shortcomings in the studies.

Part 1 indicates that, in general, lighting and crime experiments have not been planned and executed well enough for the results to be considered reliable. There is evidence of systematic bias towards a beneficial effect of lighting on crime. Industry and company funding is known to be associated with an unjustifiable excess of findings beneficial to the industry or company concerned. In some or even the majority of cases, areas with substantially elevated crime incidence (crime hotspots) have been chosen for experimental treatment. Regardless of how ineffectual the treatment might be, reduced crime afterwards relative to control areas is naturally more likely and the increased lighting tends to get the credit for the beneficial result. Pooling results to get apparently greater precision and reliability does not eliminate such systematic bias. Results obtained by an instance of this method were shown to be improbably large, given the possibility that the experimental treatment could be repeated many times, with claimed crime reductions for each treatment compounding to values far lower than have ever been observed. The main conclusion of Part 1 is that increased outdoor lighting has not been shown to have a reliably beneficial effect in reducing crime.

Marchant (2003) examined the statistical methods used in the Dudley and Stoke-on-Trent lighting and crime experiments and in the meta-analysis described in Part 1. He found several errors that were serious enough for the conclusions of these studies to be rejected. This is independent support for the writer’s findings about these studies, and it strengthens the case that the main conclusion of Part 1 is correct.

Strident campaigns based on the apparently erroneous results have encouraged public lighting projects for crime prevention in the last few years, especially in the UK. Street crime in the UK rose by 28% in the year ending in April 2002 (Hoge 2002).

New terminology was defined in Part 1 to assist in understanding the lighting and crime relationship. By its direct physical effect on visibility, light may aid or hinder criminal acts at night at the time of actual or intended commission. Examples are seeing where to break in, and being seen breaking in. Indirect effects of light act through intervening social factors, generally with time delays, and may influence crime by day as well as at night. Existing mainstream use of the term positive correlation is retained for the case where increased lighting is associated with increased crime. The contrary use sometimes found in the crime prevention literature is avoided in this work. The terms ‘adverse’ and ‘beneficial’ are used here as they are unambiguous.

2. GROWTH IN LIGHTING AND IN CRIME

Whatever the relationship, if any, between outdoor lighting and crime, it seems reasonable that the way these and related quantities have varied over the last century or more should be known and understood as a highly desirable precondition for studies of the present situation. As it happens, the required historical data tend to be sparse, beset by discontinuities and other inconsistencies, and difficult to track down. For many countries, the records for recent years are hardly any better, if they exist at all.

For periods of up to a year or so used to date in the ‘before’ and ‘after’ phases of the field-based studies reviewed in Part 1, it appears that relatively small increments in outdoor lighting have not been reliably demonstrated either to decrease or increase actual crime. Here, changes occurring over much longer intervals in lighting and crime are examined.

Ramsay and Newton (1991, p 12) commented on the steep rise in crime in the UK following WW2 while major street lighting improvements were taking place. As they put it, this “scarcely suggests that street lighting improvements are of great importance in preventing crime”. The co-occurrence of the increases suggests more than this, viz that there is a positive (ie adverse) correlation between lighting and crime. If there is a persistent correlation, then it could be useful to know why.

Others have noticed that urban crime rates and outdoor lighting both increased substantially during the twentieth century. Perhaps the notion of a connection has been dismissed each time at the outset as contrary to common knowledge that lighting reduces crime, and therefore fanciful. Regardless, the issue seems to be worth another look now that the best short-term studies have been exposed as rather inconclusive or unconvincing. Time-series measures of lighting and crime need to be examined for evidence of association and causality. Availability of reliable data determines how far back such assessments can be taken. Here, attention is limited to periods starting at least a decade after the initial introduction of electric lighting.

Many older people are able to recall that their childhood hometown skies had far more stars than is now the case. There is no doubt that their loss is a result of a great increase in outdoor lighting, but it is also important to know how this change has progressed. If the forms of long-term increase for lighting, crime and other potentially important factors such as the economy are sufficiently different, it may allow disentangling of possible causality.

2.1 Outdoor Lighting

2.1.1 Brief history

Prehistoric campfires provided warmth, light and a degree of safety at night against attack by wild animals. The flames doubtless acted also as beacons for late-returning hunters and, rather less beneficially, for hostile warriors from other tribes.

Fixed urban outdoor lighting at night was instituted in Paris of the 1660s by Louis XIV as a supposed crime prevention measure (D’Allemagne 1891). Candles gave way to oil lamps in the eighteenth century, and coal-gas lighting was introduced in London in 1809. Principal street intersections in Los Angeles were first lit by the Los Angeles City Gas Company on 1865-04-14. Edison’s incandescent carbon-filament light bulb was invented in 1879. These lamps and the electrical system to supply them were first marketed in New York in 1881, with safety as a selling feature. The first electric streetlight[4] in Los Angeles became operational on 1882-12-31 (Anon. 2002a). However, it was not until the metal filament displaced the carbon filament in the 1910s that artificial light had become sufficiently bright and affordable to start having a broad impact on outdoor activities at night.[5]

Public lighting increased greatly during the twentieth century, not only in terms of the area lit but also in the strength of illumination. For example, in an Iowa town in 1895, public artificial lighting was not required between midnight and 5 am and not at all if there was enough moonlight to see by (IDA IS114 1996), say about 0.1 lux. Between 1925 and 1996, the recommended average for street lighting in the USA went from 3 lux to 14 lux. On this sparse indication, the illuminance of outdoor lighting increased by a factor of 140 times over a century. If the growth rate in this time had been uniform, it would have been an increase of about 1.6 times per decade. Note that this takes no account of the growth in the area illuminated.

In many countries, past records of extent, use and total energy consumption of outdoor lighting seem to be limited or no longer in existence. Some historical records of electrical power generating capacity are extant (eg Hirsh 1999) but the energy fractions used for outdoor lighting, let alone the quantity of indoor lighting contributing to outdoor ambient illumination, appear far less likely to be available with the degree of regional detail and continuity required for research purposes.[6] In Australia, the inadequacy of even relatively recent records became apparent when compliance with the Kyoto Protocol was first being considered. It took some years for the government authority involved to publish total electrical energy usage for the 1990 base year of the Protocol. The outdoor lighting component for 1990 still does not appear to be available from this source, let alone for earlier years. Government departments in the UK are likewise unaware of basic facts about the amount of energy used and wasted by outdoor lighting (CPRE 2003 p 17).

Presumably, the preservation of energy use details for outdoor lighting over decades has not been seen as a priority let alone a necessity in the industry. The situation appears to have deteriorated further with the trend in some countries, including Australia, for government-owned electricity utilities to be privatised. Neglect is bad enough, but on top of this, the continued survival of industry records of previous lighting systems and energy usage seems to be at risk when community and environmental groups seek to quantify their complaints about obtrusively relit commercial premises or sports grounds, or about larger-scale developments.

Regardless of the apparent dearth of useful numerical data on the growth in total light radiated by outdoor lighting during the course of the twentieth century, there is no doubt that outdoor lighting has grown from virtually nothing to the present unprecedentedly large values. As a more or less constant minimum proportion of a city’s total outdoor light flux is scattered by the atmosphere back towards any particular point on the ground (eg Fisher 1993), measurements of this light (artificial skyglow) are a valuable independent source of information about the rate of change in the total amount of light in the city.

The absolute quantity of light radiated upward at any time can also be derived from skyglow measurements at that time or vice versa, albeit with a lot more effort (eg Garstang 2000a, Falchi and Cinzano 2000). Periods in which observational data are of poor quality or non-existent can be dealt with satisfactorily by interpolation, assuming monotonic growth tied to the known initial and observed recent values. This is especially useful for periods covering the introduction of electric lighting as the natural skyglow value is reasonably well quantified and the initial artificial skyglow value is effectively zero.

Incandescent lamps and their quartz-halogen derivatives are still in common use for outdoor lighting although increasingly they are being supplanted by lamps using an electrical discharge in low- or high-pressure gas. Relative to incandescent lamps, these newer types tend to be between four to ten times better in turning electrical energy into light (eg IDA IS115 1998). The successive introduction of improved lamps must have caused the total amount of outdoor light flux to grow more quickly than the growth in electrical energy used for making this light (eg Cinzano 2000d, Hänel 2000). The development of solid-state light sources has now reached the stage where they also are increasing the range and affordability of outdoor lighting, thereby further promoting the use of artificial light at night.

Historically, motivating factors that have led to more and brighter outdoor lighting include:

· facilitating pedestrian mobility and vehicular traffic at night by reducing the likelihood of accidents and aiding wayfinding,

· allaying the fear of crime at night,

· attracting individuals to commercial venues such as bars, theatres and shops,

· the increasing popularity of outdoor sports at night for players, spectators and television audiences,

· individual, corporate or national self-interest or self-aggrandisement, as with bright lighting of power stations, illuminated advertising and architectural floodlighting,

· increasing luminous efficacy of lamps, giving more light for the amount and cost of the power consumed,

· safety and productivity in outdoor industrial operations at night,

· promotional incentives by operators of thermal power stations to meet the base load required at night for profitability, and

· as an intuitive countermeasure to actual crime.

General experience is that results seem to confirm expectations. In some cases, especially the last two or three, however, most individuals have neither the analytical skills required nor access to quantitative data to make reliable judgements of the effectiveness of outdoor lighting. As indicated in Part 1, even the experts have difficulty in the case of actual crime.

Regardless, many practitioners of Situational Crime Prevention[7] appear to believe that more and brighter outdoor lighting at night does reduce crime, and often recommend increased lighting accordingly. The belief has permeated academia to the point where discordant evidence has caused surprise.[8]

2.1.2 Indications of growth in lighting

Pike (1976) and Berry (1976) reported skyglow growth in southern Ontario as about 7% to 10% a year. A steady rate of growth gives an exponential increase.[9] Cinzano (2000d) summarised observations in southwestern USA: 5% to 6.2 % per year increase over San Jose from 1948 to 1978, and 10% to 15% per year over Tucson prior to 1972. Shaflik (1997) referred to International Dark-Sky Association material that put the growth rate for artificial skyglow over some American cities at 30% per year (14 times per decade). For Akita City in Japan, the increase from 1993 to 1996 (Isobe and Hamamura 1998) was 16% per year, equivalent to 4.6 times per decade. Van den Burg (2000, p 33) quoted what appears to be a decadal increase of 15 times in skyglow over The Netherlands.

Cinzano (1995) summarised measurements of the astronomical V-band brightness of the night sky at Asiago in northern Italy: the value in 1993 was “at least doubled with respect to the sixties.” The V-band measure is effectively made with light passed through a green filter and is approximately proportional to luminance.[10] Allowing for the presence of natural skyglow in the measure would indicate more than a doubling in the artificial component. Since then, Cinzano (2000d, 2002) has produced a graph of Italian observatory measurements of artificial skyglow from 1960 to 1994. The trendline appears to be exponential with a decadal growth rate of about 2.6, representing an annual growth of 10% or a factor of 45 times over 40 years. For Italy overall, exponential growth in the artificial component at sea level from 1971 to 1998 was typically 7% to 10% a year (Cinzano, Falchi and Elvidge 2001b) or 2.0 to 2.6 times per decade.

A decade of satellite measurements of optical energy radiated from Earth’s populated areas (Cinzano, Falchi and Elvidge 2001a)appears to indicate a generally exponential growth of skyglow. Worldwide, only a few areas such as central India and some cities in Russia have shown a reduction in upward light emissions in recent years (NASA 2000, Sutton 2002). Satellite data for 1993 to 2000 has been analysed to determine that the total skyglow over the UK increased in this time by 24% (CPRE 2003).

2.1.3 Skyglow, outdoor lighting and population

With a moonless clear sky and well into the dark hours, the absolute values of both natural and artificial skyglow are dependent on geographical position and direction of observation. For the present purpose, the exponential growth in the artificial component added to the value of the natural component is a key issue. The luminance of the natural component varies according to the phase of solar activity. Garstang (1989a) used 53.7 nanoLambert (nL) (171 µcd/m², or 0.171 mcd/m²) as the minimum, occurring at solar activity minimum, and 55 nL for the same quantity in Garstang (2000). Cinzano (2000d) allowed a 1 stellar magnitude[11] increase over the minimum natural skyglow at solar maximum, which would give a value of 0.43 mcd/m² and a logarithmic mean of 0.27 mcd/m². The value used by Cinzano, Falchi and Elvidge (2001b) as typical was close to the latter. The amount of artificial skyglow in remote rural areas can still be much smaller than the minimum natural value. In the total skyglow over large cities, however, the natural component, even at its maximum, generally became negligible decades ago.

Walker’s law (eg Garstang 2000; Mizon 2002, p 65) is an empirical relationship for predicting the artificial component of sky luminance at 45º above a city with known characteristics at a known distance. A simplified version of the law is based on the assumption that the amount of outdoor lighting per member of the population is a constant. Astronomers such as Pike (1976) once hoped it to be so, but time has shown that this factor has continued to increase. Pike’s pioneering work on modelling skyglow growth in southern Ontario during the 1970s provided indications even then that the growth was exponential. His predictions about the Milky Way being blotted out in southern Ontario by 2000 have unfortunately proved correct.

Garstang (1991) calculated the effect of airborne dust on skyglow. Desert dust and volcanic dust have closely comparable effects. Dust below about 10 km altitude reduces skyglow and higher dust increases it. For the present purpose, the effect on skyglow is unimportant. The same conclusion applies to Los Angeles smog (Garstang 2000), so the effects of dust and smog generally can be ignored in this document.

Garstang (2000) calculated the growth of skyglow over Mount Wilson Observatory in California from 1910 to 1990 using a set amount of light flux emission (1000 lumen) per person from 124 cities in the Los Angeles basin. As in earlier papers, Garstang chose to exclude effects of changes in lighting technology, not because they were unimportant but to give a simpler basis for comparison of results. Within a large margin, the results were consistent with a few visual observations of stellar limiting magnitudes. The plotted curve overall is concave upward, but not to the extent implied by the actual observations of skyglow elsewhere in southwestern USA reported in the preceding section. Garstang (1989a) did consider introducing a factor to give faster skyglow growth than with population alone, but tested for this and considered that there was insufficient evidence to support it.

To estimate the threat of skyglow to national parks, Albers and Durisco (2002) calculated skyglow over the contiguous states of the USA. They also assumed that cities emit light in proportion to population. The resulting map looks like a satellite image of real city lights at night, but the calculated intensity of city centres appears deficient[12] and some cities look to be fainter than they are in satellite images. The second of these differences probably arises because the light emission per person in the USA does differ substantially between cities (see Section 5.2). Garstang’s calculation for Mount Wilson would tend to ‘average out’ the local variations.

Assuming that skyglow is a linear function only of population ignores the long-term increase in public lighting illuminances mentioned in Section 2.1.1. It also ignores the increasing amounts of light being used by retailers and advertisers, and the increasing use of decorative and security lighting. The amount of light produced for a given amount of electrical energy has risen with improvements in lamps. Because electricity has tended to become more affordable over the years, the amount of light provided for a given inflation-corrected cost of electricity has increased even more. All of this suggests that the outdoor luminous flux per person has been growing substantially. If city area grows linearly with population, then artificial skyglow can be expected to grow in proportion to city area multiplied by mean illuminance. Even if the time rates of increases in area and illuminance are both linear, the product of these is exponential.

2.2 Crime Data

Crime appears to have increased substantially in many countries during the twentieth century. Criminological studies of long-term variations in the crime rate (eg Graycar 2001) and its causes are hampered by factors such as:

· large shifts in what society regards as a crime (eg drunkenness, nude bathing, homosexuality),

· the emergence of new crimes (eg Internet fraud, aircraft hijacking),

· political and industrial influences on reporting and recording of crime and its solution rates,

· improvement in methods of detection and prosecution of offenders, and so on.

The problems of poor data quality and inadequately standardised crime definitions have been known for over a century but are still being dealt with. Differences in criminal justice systems add to the problems of comparing international crime rates (eg Barclay and Tavares, 2000). Regardless, there are often inexplicably large differences between countries in specific crime rates, eg homicide (Graycar 2001).

The following sections present within-country comparisons of available lighting and crime growth data.

2.3 Growth in Lighting and Crime in Individual Countries

2.3.1 Australia

Despite substantial effort, the writer did not find suitable records of energy use for outdoor lighting in Australia and the proportions of various lamp types installed. The same applies to records of illumination levels or total light output, so recourse has been had to astronomical observations.

For the period 1880 to 2001, artificial skyglow values shown in Figure 1 (all of the figures are grouped at the end of this document) have been inferred from limiting unaided visual stellar magnitudes assembled by the writer from observations in recent decades from Melbourne suburbs plus a few more as far back as the mid-1950s (reported in Dudley 2000). Fluctuations must have taken place during social upheavals such as the Great Depression and the two World Wars, especially WW2 because a blackout was enforced,[13] but they are not shown because of the absence of specific data for the years in question. The skyglow changes caused by population changes at those times were small in the USA (Garstang 2000), but social, economic and technological effects on skyglow could well be larger than the population effects. The sky reference area chosen for skyglow observations and calculations is at 45 degrees altitude, due south of an observing site about 11 km northeast of the Melbourne central business district. The zenith might have been a better reference position, but this is relatively unimportant for the present purpose.

The pre-electric-lighting ordinate of the skyglow curve in Figure 1 is 0.27 mcd/m², typical of the luminance of the natural clear moonless night sky. For the luminance of the Melbourne sky to start at that value and to be where it is now and where it has been in recent decades is most readily explained if the growth rate of the artificial component is taken as exponential. The observed increase is close to doubling in each decade.[14] To sufficient accuracy for the present purpose, the skyglow luminance S mcd/m² is given by

S = 0.27 + 0.00189*2^((year-1880)/10).

On the basis that there do not appear to have been large changes overall in the shading of installed outdoor light fittings against upward light spill in recent decades and that probably only small changes have occurred in the amount of light reflected upward from the terrain, the plotted skyglow curve in Figure 1 is considered to indicate how the total of artificial ambient light outdoors at ground level has grown in recent decades over the more densely populated areas of Australia. The growth rate has probably been greatest in the state capital cities, and slowest in rural areas. Determining the actual lighting growth rate in rural areas using skyglow observations is complicated by the additive effects of waste light from distant cities and nearby towns. Examples of this are given by Berry (1976), Cinzano (2000c,d) and Dudley (2000). Note that luminance, rather than log luminance, is additive.

The form of the skyglow curve in Figure 1 can reasonably be supposed to be typical for most of the highly urbanised Australian population. Of course, long runs of frequent observations of skyglow in various locations would be better to work with, but so far, few Australian skyglow observations seem to have been published.

Figure 1 also shows three sets of Australian recorded crime data. These data were selected on the basis of being available, appropriate and freely accessible through the Internet for ease of checking and comparing by others.[15] Data for homicide were also available but not included as the rate has not varied greatly over the whole century[16] (Graycar 2001). The numbers for homicide are so small relative to those for other crimes that they would make no practical difference anyway in any overall number or rate. For the crime data that are shown in Figure 1, partition into day and night occurrence would be valuable but such data did not appear to be available, either on the Internet or elsewhere.

The yearly crime rate data for 1974 to 1992 are from Walker (2002), for 1993 to 1999 from Graycar (2001), and for 2000 and 2001 from ABS (2003). The usual basis of reporting crime rate per 100 000 of the population has been changed proportionally to smaller population numbers for convenience in plotting the data with a common ordinate scale.[17] The data are not quite internally consistent and seamless because of evolving differences between state and territory jurisdictions in definitions of crime, late numerical corrections and other relatively minor perturbations. For example, data included as Burglary are called Unlawful Entry with Intent in two of the source documents. These shortcomings again appear to be of little consequence for the immediate purpose.

Part or all of the drop in motor vehicle thefts in the 1990s appears to have been a result of various measures stemming from political concern, such as increased attention to insurance fraud, vehicle ‘rebirthing’ and anti-theft devices (eg Grabosky and James 1995, Section 14).

The Australian population is also plotted in Figure 1 using data from Lahmeyer (2002). Missing data for some early years have been interpolated. Population data are shown because they are readily available and because many of the social factors that might be thought to encourage crime (eg housing density) seem likely to be related to the size of the national population. Given the different character of the skyglow and population growth curves, it was thought possible that correlations of each with crime data might indicate any relative differences in degree of association.

Australian crime records are available for six of the years in the period 1890 to 1902, but they are given only as offences against the person and offences against property. Public drunkenness is not included here as it is no longer a crime. The total rate of violence and property crime for this period in Australia fell from about 11 per 1000 to under 7 (Graycar 2001). A roughly equivalent total rate of violence and property crime in the last years of the twentieth century is over 60 per 1000, ie about 7 times greater. The total rate for all crime in Victoria was about 40 per 1000 in the first half of the twentieth century, doubling from 1945 to 1955 and trebling by 1975, reflecting an Australia-wide pattern (Walker (2002), hard copy version cited by DCPC (2002) p 11).

Over the twentieth century, the Australian population grew by a factor of about 5. The increase in natural plus artificial outdoor light flux in the Melbourne metropolitan area in that time is about 30 times, less if moonlight is included in the natural flux, or more if moonless overcast nights are considered. It is more again if the peak ambient illuminance in large city centres is considered, or less for the typical ambient illuminance in the outer suburbs. Crime rate growth (day plus night) is obviously correlated both with lighting growth and with population growth over decades. Neither appears to be a markedly better correlate than the other.

Discussion of causality would be premature at this stage. Readily available Australian crime data appear to be of insufficient quantity and quality to get much further with this line of investigation. In the absence of reliable crime data for the earlier part of the twentieth century, attention has been directed to other countries with better records.

2.3.2 England and Wales

Population data for 1900 through 1997 from the House of Commons library (Hicks and Allen 1999) for England plus Wales are plotted in Figure 2. The population graph was extended by linear extrapolation back to 1898 and on to 1998. Total crime data from the Home Office (2002b) for 1898, 1899, and to 2001 (retaining the pre-April 1998 counting rules) have also been used to extend the graph. The crime data set covers a much longer span than that for Australia in Figure 1.

Actual historical data for the growth of skyglow in England and Wales would be ideal to include in Figure 2, but not even the world’s first book in English about light pollution (Mizon 2002) includes this despite its UK origin.[18] Nor were any quantitative records found in Internet and literature searches until May 2003, when the CPRE (2003) analysis of satellite data provided an exponential growth rate for overall light pollution from the UK: 24% for the 7 years from 1993 to 2000. As an exponential rate, this is equivalent to a 3.12% growth per year or 36% in a decade.[19] To derive the skyglow growth curve for the UK starting from the 1890 natural skyglow value of 0.27 mcd/m², exponential growth was assumed, with the value at 2000 set equal to that used in Figure 1, ie 8.0 mcd/m², corresponding to medium light pollution with about 3% of the naturally visible stars still visible. To sufficient accuracy for the present purpose, the total skyglow S mcd/m² is given by

S = 0.5*1.03121^(year -1910).

This growth curve is shown in Figure 2. It is intended to be representative of the sky seen from a suitable location near any UK city or large town. The lack of account of the war blackouts and the 1930s depression detracts somewhat from its usefulness. Fortunately the scale of the skyglow curve is immaterial in ordinary correlation calculations. The shape of the curve is of more importance. As can be seen in Figure 2, its upward concavity tends to match the long-term trend of the police-recorded indictable crime rate in England plus Wales.

In Figure 2, the crime rate since 1989 appears to be a combination of reasonably steady growth plus an increment peaking in 1991 or 1992. In following sections, this feature will be seen to have approximate counterparts in several other countries. The subsequent decline back to the long-term trend may have been due entirely to change in the unknown factors that produced the increment (eg fluctuations in the supply of illicit drugs?), but the Home Office (2002a) treated the period of the decline in isolation in attributing the decline partly to the contribution of police and their partners in crime control. This is not to deny the value of police or of outcomes of the Crime and Disorder Act 1998 but is a reminder of the difficulties of interpreting short runs of time-series data.

2.3.3 USA

By itself, the existence of good correlations between lighting and crime in Australia and in England plus Wales is not sufficient to justify international generalisations from these national findings. Criminologists caution that international comparisons of crime rates can be misleading (eg Barclay and Tavares 2000). Direct comparisons from 1981 to 1996 between crime data for the USA and England plus Wales (Langan and Farrington 1998) indicate that there are substantial differences between the respective rates and trends for specific crime types. Therefore it is important to extend this investigation to other countries. The USA is a readily justifiable choice because of the apparent wealth of US crime data readily available online, some of it even dating back to 1860 for cities (eg Monkonnen 1994).

Unlike the situation in England and Wales, however, usable crime data for the USA spanning the twentieth century were not found. Most of the available time-series data for US crime only covers recent decades. Inspection of Bureau of Justice Statistics data starting in 1973 (BJS 2002a) indicated that correlations of surveyed property and violence crime rates with skyglow and population would be small or negative. However, correlations with recorded violent crime and arrest rates still appear to be positive.

Annual crime rate data for the USA from 1960 are given by Maguire and Pastore (2002, Table 3.120), based on BJS Uniform Crime Reports (UCR) data. For years 1933 to 1959, the source was JRSA (2000). The total of property and violent crime rate, including homicide, is called the Total Crime Index or the Index Crime Rate, shown in Figure 3. The discontinuity at 1959 is an artifact resulting from change to the dollar limit for larceny. The 2001 results (Butterfield 2002) are not shown in the figure: the UCR rate increased by 2.1 % over the 2000 value, ending the ten-year decline. During 2001, robberies climbed 3.7 percent, burglaries 2.9 percent, petty thefts 1.5 percent and motor vehicle thefts 5.7 percent.

No regional or national skyglow data were readily available so the skyglow curve used in Figure 1 is used again as representative of exponential growth of skyglow near cities. US population (Demographia 2002), the combined adult and juvenile estimated drug arrests rate (BJS 2002a) and the total US federal and state rate (BJS 2002a) are also included in Figure 3. The incarceration rate was included as it could possibly account for at least part of the reduced growth or falling trend of crime in the US while the various factors conducive to crime presumably continue to increase.[20]^,
[21] Over half of the steep increase in US prison population since 1990 is a result of convictions for violence.

Drug offences were included in Figure 3 as an exception to the flatter trends for most other crime in the US over recent decades. All or most of the difference in growth rates may well be caused by factors other than any effects of lighting. At first sight there might seem to be no good reason to expect any direct connection at all between lighting and drugs crime. However, exposure to artificial light at night already appears to have reached levels great enough in some cases to reduce the production of melatonin by the pineal gland (see Section 6.1.4 below). This is so both in the evening and then later in the small hours when this part of the human endocrine system appears to be most sensitive to small amounts of light.[22] If drug taking is indeed somehow affected by endocrine disruption as an explanation of the differences between drugs crime and other crime in Figure 3, this would not necessarily affect the relative day-to-night ratio for particular kinds of offences.

2.3.4 Canada, New Zealand and other countries

Recorded crime data from 1962 to 2001 are available for Canada (eg Statistics Canada 2001). Variations in the Total Criminal Code rate appear to mirror those in the US Total Crime Index shown in Figure 3, although the ordinates are higher. The recorded overall offence rate for New Zealand for years 1970 to 2001 (Statistics NZ 2002) has a somewhat similar shape and ordinates a little higher again. Note that crime rate differences between countries are not a reliable guide to relative inherent criminal tendencies because of confounding by differences in laws, policing and justice procedures, statistical treatments and other relevant factors (eg Barclay and Tavares 2002). Therefore, international differences in any environmental influences on crime should not be inferred from such crime rate differences.

The United Nations has collected annual total recorded crime rates for 88 member countries (UN 2000). The years covered are 1980 through 1986 and 1990 through 1997. Most of the responding countries have not provided data for all of these years. In several cases, the data are obviously incorrect or unreliable, with non-credible minuscule values or massive changes from one year to the next. Political instability and adverse consequences of war seem likely to have led to some of the effects.

National data were selected from the UN table with the following criteria: data runs had to cover at least 12 years apart from the three-year gap from 1987 through 1989. In one case, Argentina, missing data for 1993 and 1994 were estimated by linear interpolation between 1992 and 1995. All missing data for 1987 through 1989 were estimated by linear interpolation. Countries with apparently incorrect or unreliable data were excluded, unless the problem was a single typographical error such as a factor of 10 in one entry for Japan. Small countries were not included, nor were the five countries already considered (Australia, Canada, England plus Wales, New Zealand, and the USA). This left 15 countries. Their data are shown in Figure 4. All show a positive gradient overall. The mean linear increase in crime is 4.5% per year (range 1.08% to 11.3%). This is a factor of about 1.5 per decade. The crime growth rates in the five countries already considered are less than half of this. Nevertheless, it does indicate that any association between artificial light at night and crime does have global applicability instead of being confined to, or most active in, countries with some common feature, such as English as the principal language.

2.4 Correlations Between Crime and Other variables

Linear correlation coefficients calculated for the crime and justice data in Figures 1 to 3, and for similar data from Canada and New Zealand, are shown in Table 1, below in this section. The values are generally high but do not indicate a consistent superiority for either lighting or population as a correlate with crime. The balance seems to shift a little towards lighting if crime and justice numbers are used instead of rates. The change to numbers can be achieved by multiplying rates by population. This has the same effect on correlation coefficients as leaving the rates as they are and dividing skyglow by population, which gives a variable proportional to the amount of outdoor artificial ambient light per person. This variable is mentioned again in Section 5.2.

It seems unlikely that the picture presented by these correlations would change much in either direction if data for more of the whole century became available. By inspection, Garstang’s population-based California skyglow growth data would appear likely to give correlation coefficients with crime similar to those for the population. Garstang’s data were not included in the figures or in Table 1 because the observational evidence for exponential growth is so strong.

Hays (1970, p 565) stated, “There is no guarantee that all psychological relationships of theoretical or applied interest must be linear in form.” The logarithm of light flux might be a more suitable variable to investigate for linear correlation, given that the visual system tends to respond to light in a logarithmic fashion, and given the apparent exponential nature of existing growth in artificial skyglow. Hays (1970, p 563) indicated that using the logarithm of an exponential quantity in this way is appropriate. The correlation coefficients of the various crime and justice data, and population, with log skyglow are therefore included in Table 1. Simple counts indicate that these log skyglow correlation coefficients with crime and justice data are most often the highest values across the three columns. The small numbers and partly overlapping data sets deter a formal significance test but future work might be better organised to deal with this issue. Not surprisingly, population correlates better with log skyglow than with skyglow in 4 of the 5 cases in Table 1.

A further caution seems worthwhile about the relative amounts of crime in the countries studied. The crime data are from police records. Crime surveys in 14 countries indicated rather different relativities, and also revealed a latitude effect- colder countries tended to have lower crime rates (Walker, Wilson, Chappell and Weatherburn 1990). However, it is the rate and direction of change of crime, and of artificial lighting, with time that are of most interest here, rather than the absolute scales of crime and lighting.

It might seem rather naïve to expect total crime to correlate strongly with lighting, given that crime was problem long before bright lighting became widely available or used[23] (eg Sharpe 2002). Therefore, it might appear more reasonable to use the increment in crime above some pre-lighting base crime rate as a variable in looking for lighting and crime correlations. Subtracting a constant base rate from any of the crime data does not change linear correlation coefficients, however. Likewise, it is almost immaterial in this context whether total skyglow or just its artificial component is used. The correlations shown in the table are a reasonable guide as they stand. The squares of the correlation coefficients have been included to indicate the proportions of variance in crime and justice data explained by skyglow and population.

Correlation studies rate a mere 1 on the Scientific Methods Score used by Sherman, Gottfredson, MacKenzie, Eck, Reuter and Bushway (1997), but that is in the context of experimental hypothesis testing rather than exploratory analysis of complex systems. The correlation studies used here are quite appropriate for guiding hypothesis formulation. Furthermore, the data sets used in this case are massive, applying to national populations over three to ten decades.

TABLE 1. Correlations of Crime, Justice, Population and Skyglow

Country

Quantity:

Crime Rate,

Justice Rate,

or Population

Inclusive

From To

Linear Correlation Coefficients r (above, r² below) with

Skyglow

Log

Skyglow

Population

Australia

Motor Vehicle Theft

1974 2001

0.626

0.392

0.759

0.576

0.770

0.593

Robbery

1974 2001

0.985

0.970

0.945

0.893

0.937

0.878

Burglary

1974 2001

0.873

0.762

0.953

0.908

0.954

0.910

Population

1890 2001

0.858

0.736

0.988

0.976

England

plus Wales

Indictable Crime

1898 2001

0.970

0.942

0.870

0.757

0.826

0.682

Population

1898 2001

0.899

0.808

0.995

0.989

USA

Recorded Total Crime

(UCR Index)

1933 2001

0.683

0.466

0.895

0.801

0.913

0.834

Recorded serious violent crime (not shown in Figure 3)

1973 2001

0.622

0.387

0.764

0.584

0.750

0.569

Estimated Total

Drug Arrests

1970 2001

0.926

0.857

0.926

0.857

0.925

0.856

Incarceration, Federal plus State Prisons

1980 2001

0.980

0.960

0.995

0.990

0.995

0.990

Population

1900 2001

0.989

0.978

0.999

0.998

Canada

(Not shown in figures)

National Criminal Code Rate

1962 2000

0.626

0.392

0.821

0.674

0.808

0.653

Population

1890 2001

0.835

0.697

0.821

0.674

New Zealand (Not shown in figures)

Overall Offence Rate

1970 2001

0.758

0.575

0.889

0.790

0.824

0.679

Population

1970 2001

0.980

0.960

0.983

0.966

For each of the three countries studied in Figures 1 to 3, real expenditure on police has increased over recent decades. For example, allowing 1.726 for the total inflation, the real per capita expenditure increase for all US police forces combined over the period 1980 to 1999 (BJS 2002b) was a factor of 2.016. The bulk of this expenditure is likely to have gone into increasing police numbers rather than into more and better equipment and training (Greenberg 1983).

Doubling of police numbers represents a compounding 10% increase repeated about 7 times. Using Marvell and Moody’s (1996) estimate of the effect of a 10% increase in police numbers, a doubling of police would reduce crime to (0.971)⁷, ie 0.814, or a reduction of 18.6%. Goodman’s (2002) first estimate of the same effect is more appropriate here than his second estimate. The first estimate indicates a reduction to (0.989)⁷ or 0.925, ie a reduction of 7.5%.

Had there not been progressive real increases in police force expenditure in the countries studied, some notional progressive increases would have occurred in the crime rates instead. The overall effect of this would again be to improve the correlation with skyglow a little more than with population. In the case of the USA, if real police expenditure had remained at the 1980 level, the UCR crime rate value for 2001 would have been nearly 30% higher than is shown in Figure 3.

Australia, UK and USA all increased their prison populations substantially in the last decades of the twentieth century. As with the effect of police, criminologists differ about the effect of incarceration on the crime rate. Even if the deterrent effect of incarceration is discounted, the balance of opinion seems to be that incapacitation of offenders does have a substantial effect in reducing overall crime. Langan (1994) concluded that the increase in US prison population from 1975 to 1989 reduced reported and unreported crime by between 10 and 15%. Hayward and Izumi (1996) collected estimates by others of how many serious crimes per prisoner per year are avoided by incarceration: the range was 12 to 187. Marvell and Moody (1994) reached a conservative estimate of 17 and a more likely value of 21 crimes prevented per prisoner per year. The effect of this on the crime rate is like that of increased police resources: in its absence, crime would be worse than it is and again the effect would be a steeper rise in recent years and a better correlation with lighting.

Using Marvell and Moody’s value of 21 crimes per prisoner per year, had their been no increase in incarceration in the USA since 1980, the UCR crime rate value for 2001 would have been nearly 70% higher than is shown in Figure 3. Adding this to the effect of no extra police, the net result would be a 2001 crime rate value about double that shown in Figure 3. The case for a high correlation of lighting and overall crime is a lot stronger than Figure 3 indicates.

In future work, it may be useful to have a variable representing the tendency of a population to criminality. It would mainly depend on the reported or surveyed crime rate, but would include an increment that would apply in the absence of a police force (or perhaps in the presence of a minimal police force to avoid an implied state of anarchy) together with another increment representing the extra crime that would have occurred had no criminals been incarcerated.

Empirical results of modelling by Witte and Witt (2000) suggested a long-run equilibrium relationship between crime, prison population, female labour supply and durables consumption. This suggests the possibility of using outdoor lighting measures such as skyglow or ambient illuminance at night in multi-factor models, along with or in place of durables consumption.

2.5 Crime by Day and Night

Time-series crime data are common. Data differentiating crime by time of day are less common. The combination of these appears to be quite uncommon. Reported burglary data for the USA (Maguire and Pastore 2002, Table 3.163) are in this category and provide the time-series percentages shown in Figure 5. Clearly, the percentage of night burglary has fallen steadily relative to the percentage of day burglary over the 25-year period. The total annual number of burglaries has also been falling steadily. Because of the general increase in lighting over this time, these data might be thought to favour an interpretation that lighting is more of a hindrance than help to burglary at night, at least in the USA. However, it would seem illogical to claim that increasingly bright artificial lighting is displacing burglary to the brighter conditions of daytime.

Other plausible explanations exist, including supposed or known increases over time in:

the deployment of police at night,

· private security patrols at night,

· residences unoccupied during the day, because of:

· trends towards smaller number of occupants per residence,

· workforce participation by both parents of young children,

· number of intruder alarms installed and a presumed tendency for these to be switched on more at night,

· occupancy of workplaces at night, and

· drug-related burglaries and adverse side effects of drugs on vision, especially in dim light.

In some jurisdictions there are heavier penalties for unauthorised entry with intent into houses with an occupant present, or at night, or both. It is not known if these laws and penalties or changes to them have had any material influence on the present issue.

Figure 5 presents some further lessons. Non-residential burglaries happen more at night than by day, although the difference is declining. Residential burglaries have been more common by day than night since 1976 and the difference is increasing. During a typical 24 hours, this means that the rate of burglaries must change as the light changes from the night level to the day level. For residential burglaries, it rises as the light increases and falls as the light decreases. The opposite happens with non-residential burglaries, but in smaller numbers and the crime is much less feared by most people. Neither case implies causation by light, just as the shift in proportions over the years does not imply causation by light. Some factors other than light and lighting, by themselves, must be having a substantial effect on the day-night ratios.

As will be demonstrated in following parts of this document, there appears to be no known reliable evidence that lighting, by itself, has a net hindering effect on burglary at night.

Lighting appears to provide net assistance in the commission of graffiti and possibly other vandalism, crimes that have much in common with the surveillance, approach, break and escape phases of burglary. External light would appear to be generally advantageous to criminals, especially for the indoors part of burglary at night. Of course, the trends in Figure 5 may just reflect a growing belief or experience among criminals that burglary in broad daylight is more likely to succeed, regardless of the reason and whether the belief is true or not.

3. DARKNESS AND CRIME

3.1 Dark Campus‘Dark Campus’

As vandals apparently need or like to see the damage they cause,[24] this led to the notion that making areas dark could actually reduce vandalism (California Energy Extension Service (no date), King 1995, Wilson 1998, p 8). The San Antonio School District in Texas was one of the first to try the Dark Campus idea in 1973. All of the lighting in the area was turned off in an around each of the 19 schools after they closed for the night. The annual cost of repairing damage caused by vandalism dropped from $160,000 in the late 1960s to $41,000 in a few years, and annual energy savings amounted to hundreds of dollars per school. The following statement is attributed to the Director of Security for the school district:

“I remember as a kid, we never hung around in the dark. We hung around a street light or some other kind of illumination. We wanted to see who was with us. With vandalism, the thrill is seeing the windows broken, in seeing the words written on the wall. It is no thrill to hang around in the dark”. (California Energy Extension Service, no date)

In California, the Livermore Joint Unified School District reported annual energy savings of about ten percent after introducing a dark campus policy in 1974. The Director of Facilities Maintenance is reported as saying:

“A dark campus policy positively will not increase vandalism. This is what many people are concerned about, including the police, but it did not happen. We’ve noticed a slight decrease in vandalism over the years, but we have done other things too, like vandal watches and Sonitrol.” (California Energy Extension Service, no date)

Cupertino Union School District reported that vandalism dropped 29 percent with an anti-vandalism strategy including a campus blackout, while energy savings totaled $8,190 during the 1981-82 school year. Battle Ground School District in Clark County reduced vandalism to almost zero with a policy to darken campus after 10-30 p.m. Spokane School District and Riverside School District have been experiencing similar results for over six years (California Energy Extension Service, no date).

A retired Associate Superintendent for the East Side Union High School District in San Jose, CA, is reported as saying:

“We are not aware of any school districts where blacking-out campus coincided with an increase in vandalism, burglary, or arson. There has not been an increase of such incidents in our district during the hours of total blackness. It seems logical that a blackout discourages youth from entering campuses -- they have as much fear of the unknown as anyone else. In case of burglars, any light shown on a campus is cause for suspicion on the part of neighbors and police.” (California Energy Extension Service, no date)

Eight schools in Eugene, Oregon introduced Dark Campus in 1989. Vandalism virtually disappeared in certain problem schools. Stanley (1993) stated that turning off all lighting when school is closed saved nearly $14,000 in six buildings of a small Illinois school district and over $150,000 annually in 19 of 32 buildings in another Illinois district. Vandalism and loitering were claimed to have decreased. Success has also been claimed for the dark campus approach in Tampa, FL (Hollingsworth 1995) and Tucson, AZ (Wilson 1998). Marion County Public Schools (2000) (USA) reported a study of costs of vandalism and the potential saving in power costs that would follow adoption of a Dark Campus program.

Roller (2001), a maintainer of outdoor lighting fixtures at a US university, claimed that vandalism increased almost immediately where lights had failed. No data accompanied this assertion. He also expressed dismay that several previous letters in the same magazine had suggested that darkness deters crime, and then stated that “Studies have consistently shown that proper lighting greatly reduces the risk of all sorts of crime”, a claim that is refuted by the evidence presented in Part 1 and in this document.

Blakemore (2001), reporting on the US ABC News without giving specific details, mentioned large savings by switching the lights off in office and school buildings. He stated:

“Police report that such darkness is often safer. That’s partly because neighbors soon learn to alert police if they see any lights on in a building. There’s even less graffiti because it’s usually lighted walls that attract the spray-can vandals, not dark ones.”

The last sentence is consistent with experience in Melbourne. The underground rail tunnel is lit at intervals and is dark in between. Colleagues of the writer have observed that graffiti is strongly concentrated at the lit parts of the walls.[25]

More material on campus lighting and crime is in IDA IS23 (1996), IDA IS27 (1997) and IDA IS31 (1997). Note that concurrent anti-vandalism actions could confound analysis of the effectiveness of darkness. At 1997, the success of dark campus programs in combating vandalism had not been backed by scientific evidence (IDA IS54 1997), but the California Energy Extension Service (no date) was not aware of any reported increase in vandalism due to night time blackouts.

The available evidence presented in this section suggests that darkness inhibits property crime and that the effect in the circumstances applying is primarily direct.

3.2 Large-Scale Darkness

3.2.1 Auckland

Pease (1999) quoted Fennelly (1996), who asked:

“What would happen if we switched off all the lights at night? … Such a foolish act would create an unsafe environment. Senior citizens would never go out and communities would have an immediate outbreak of thefts and vandalism. Commercial areas would be burglarized at an uncontrollable rate. Therefore, lighting and security go hand in hand. The above example may seem to be far fetched, but in fact installation of improved lighting in a number of cities has resulted in the following: [decreases in vandalism, street crimes, suspicious persons, commercial burglaries and a general reduction in crime].”

Pease noted that no evidence was given in support of these assertions.

As Fennelly was writing in a security handbook of which he was also the editor, and did not cite any experiment that reliably showed these results, he was using the method of authority as well as the method of intuition (eg Martinez-Papponi 2000) rather than scientific method. The supposedly far-fetched actually happened because of management shortcomings less than two years later in Auckland, total population about 1.2 million and New Zealand’s largest city. Failure of high voltage underground cables supplying the central business district began on 1998-01-22 but a muted reference to this was made public only on 1998-02-19. All four cables had failed by the next day, causing a power loss that affected the city for several weeks (Maps Group 1999). Power was back to normal by end of March but it was not until mid-April that the system as a whole was operational to the extent that back-up generators were no longer required (Donnelly 2002).

Sixteen days into the disruption, widely circulated press reports stated:

“Even criminals have deserted the darkened streets of downtown Auckland… ‘It’s been almost a crime free zone’ Inspector John Mitchell said… ‘The normal level of muggings, violence, fights, burglary and robbery have just not happened.’ ” (eg Kiwi World 1998, ICOLE 2002)

Note that Pease’s review was published in 1999 and its separate summary in 1998. Neither refers to this compelling demolition of the belief that darkness and crime go hand-in-hand. However, websites of many crime prevention organisations, even in New Zealand, still emphasise their belief in the efficacy of lighting to overcome darkness and thereby to prevent crime!

In the present context, the Auckland power loss is sufficiently important for some elaboration. Initially the civil defence organisation was activated, unnecessarily, but an emergency was not declared (Davis 1998). Police moved in to the central business district to control the expected crime wave, leaving other areas with a lack of policing (Gutman 1998). The Mayor also told people to stay out of the central business district, but the streetlights were on and a night parade went ahead successfully as scheduled on the periphery of the central business district. Within a few days the Mayor and police realised there was no threat of a crime wave and no need for people to keep out of the central business district, and shopping was encouraged to help hundreds of struggling small businesses to survive (Donnelly 2002).

The Auckland central business district covers about three square kilometres, has a residential population variously claimed to be 5000 or 6500 and about 59 000 or 68 000 workers commuting to 7000 businesses. Customers and students add many more thousands (Davis 1998, Maps Group 1999, Donnelly 2002). Auxiliary power of up to a quarter of the normal load was eventually available via old cables (Gutman 1998). Because many electrically operated doors had to be propped open or left open to allow access, there was a heavy demand on security services (Gutman 1998).

Some traffic lights and streetlights were off intermittently. Domestic and retail lighting and illuminated advertising signs were generally off. There were virtually no retailing, hospitality or entertainment customers, day or night. Many residents moved out to friends’ homes (Davis 1998). One bomb threat against the power company was taken seriously (Gutman 1998).

Before-after photos on the Internet showed Auckland in its normal lit state, with floodlit buildings, glary lights and urban skyglow, compared with its much darker limited electrical power condition. It was not completely dark as there was some skyglow and light trespass from surrounding city and suburban lighting providing a low level of ambient light, together with local lighting from many of the streetlights still operating from emergency power supplies in the central business district. But crime was almost non-existent![26]

As New Zealand police statistics for 1998 were no longer available on the police website, Auckland police district crime data for March and April of 1997, 1998 and 1999 were kindly supplied by the police on request. The data were not differentiated by time of day. There is no discernable effect, increase or decrease, of the power cuts on the recorded total crime. Subsequent correspondence indicated that the downtown area mentioned by Inspector Mitchell was only a small part of the total area covered by the data. This would explain part of the apparent discrepancy between the police report and police data. The remainder of the discrepancy could indicate spatial displacement of crime from downtown to elsewhere within the police district, or early redeployment of police to these other places where they were available to make more arrests there than usual, or both of these.

The apparently immediate reduction in downtown crime following the power failure suggests a direct effect, but its continuation over subsequent days as well as nights indicates that an indirect effect also applied. The immediate and sustained effects on commerce could likewise be considered as a mix of direct and indirect effects. The significance of this will become apparent in Section 4.4.2.

3.2.2 Massachusetts

The Auckland experience is sufficient to show that Fennelly and Pease are both quite wrong, although an objection could be raised that this one blackout could have been a unique aberration. However, there are other known blackouts and brownouts. Answers to Fennelly’s question were already available on the Internet at the time he was writing, and they certainly differ from his expectation. For example, early in the 1990s, the towns of Bernardston and Northfield in northwestern Massachusetts turned off most of their street lighting for budgetary reasons. Each year thereafter, the police chiefs in both towns reported no increases either in traffic accidents or crime as a consequence of the lighting reduction (NELPAG C12 1995). This was still the situation two years later (NELPAG C14 1997).

Since then it has been reported that “numerous other towns in Massachusetts have turned off their lights with no increase in crime or traffic accidents” and that the Boston suburb of Lexington (pop 28 000) had turned off every second streetlight out of the original installed total of 3600. Complaints were resolved by turning individual lights back on. The complaints had virtually stopped by the time that the number of streetlights left off had been reduced to 1300 over several months (Green 1997). About 40% of the original number were still not operating ten years later in year 2000, and the Town of Acton had applied a similar money-saving constraint with a 20-year moratorium on additional streetlights (MMA 2000). Lexington has also consolidated various other existing outdoor lighting constraints into a single bylaw (Bowyer 1998).

3.2.3 Europe

Because upward waste and used light could provide precise information to bomber aircraft about the identification of targets at night during World War 2, the Blackout was imposed in Britain from 1939-09-01 to 1945-04-30, although a relaxation to ‘dim-out’ conditions was allowed in the last six months. Before that, shutters, black paper, cloth or paint had to be used to block the egress of light from all windows and skylights. Severe penalties could be imposed even for small light leaks. All streetlights were turned off. Vehicle headlights were fitted with slit apertures. Even the glow from a lit cigarette outdoors was regarded as a contravention.

People complained bitterly that the Blackout saw crime rocket, especially petty crime such as pocket picking and the raiding of vegetable patches. Crime did increase but people exaggerated (HSHF 2002). No evidence has been found by the writer to indicate that the imposed darkness rather than the stringency of war was responsible for changes in the amount or type of crime. Crime is now known to increase in the vicinity of military camps (Goodman 2002), and there were a great many of these present in Britain at the time.

More recently, Pearce (1995) claimed that in West Sussex (UK), police found a reduction in crime rate when streetlights were switched off after midnight.

Crime has an urban concentration (eg Graycar 2001). Details for a part of London (Camden Community Safety Partnership 2001) indicate that the concentration varies with the type of crime and the nature and function of the built environment. It has been widely accepted that the built environment has an influence on crime, but the Auckland experience in 1998 appears to indicate that the presence of light at night is more important.

Not much information is to hand about other parts of Europe. It is known that power failures affecting outdoor lighting have occurred. Nations and large regions have had street lighting curfews after 11 pm or midnight for decades in the past. Van den Burg (2000, pp 34-39) contains some information about public and sports lighting curfews in The Netherlands. There is scope for some useful research in relation to crime and changes in street lighting hours. Likewise, the recent introduction of outdoor lighting controls in nearly half of the Italian regions and in Czechia will provide further opportunities for studies of the social effects, including the usefulness or otherwise of differentiation into direct and indirect effects.

3.2.4 Great Northeast Blackout of 1965

BHP (2000) and Cooney and Stone (2002) described the power failure (outage) that affected eight states in the north-eastern parts of USA and Ontario province of Canada from about 5-30 pm Eastern Standard Time (1730 EST) on 1965-11-09. New York City’s blackout was the longest, up to 13 hours in parts. There were many minor traffic accidents, presumably because of the failure of traffic lights, but no catastrophes or disastrous fires. There were no crime waves or looting sprees. Crime rates in the region were actually well below normal. In New York City, “… criminals stayed home” and there were only 96 arrests all night (Newsweek 1977, pp 18, 24). The usual number was about 600 arrests in 24 hours (Corwin and Miles 1978, p 60).

Haas (2002) from Connecticut, reported:

“I remember that blackout quite well. The whole northeastern grid blew out and we were left without power for three days. Reporters indicated similar stories about crime throughout the CT area -- the statistics dropped dramatically for violent and non-violent crimes during a period when the race riots were common in the northeast. The following month (the original) Life Magazine did a several page spread on NYC telling how not having any electricity brought everyone closer together during times of crisis...

The most profound thing I recall was how detailed the full Moon looked to the naked eye when it was near the horizon. It was amazing and lots of fun living by candlelight for this 12 year old kid at the time. We had absolutely no trouble seeing outside at night either, thanks to the Moon.

Similar reports of no crime during blackouts occurred again during the Los Angeles earthquake in the mid-1990s when electricity supply lines were severed. Living on Earth did a feature story on Torrance Barrens preserve in Canada in 2000 that mentioned this event and told of people in LA being terrified by a strange grey cloud that split the sky in half. Many people thought it had something to do with causing the quake, but alas, it was only the Milky Way they were seeing -- our stellar back yard in the Universe. A link to that audio and related text appears in the Articles section of the LiteLynx List for anyone interested in reading it.” [See LiteLynx (2003).]”

The following event times for New York City were obtained from USNO (2003).

For 1965-11-09:

Sunset 1643 EST

Moonrise 1712 EST

End of Civil Twilight (Sun depression 6°) 1712 EST

End of Astronomical Twilight (Sun depression 18°) 1817 EST

For 1965-11-10:

Start of Astronomical Twilight 0503 EST

Start of Civil Twilight 0608 EST

Sunrise 0637 EST

Moonset 0809 EST

Full moon had occurred on 1965-11-09 at 0435 UT (BAA 1965), which is 2335 EST on the preceding day. The moon was therefore about 0.7 days past full and above the horizon when the blackout started. Thirteen hours later, when the last of the blackout apparently finished in the city, the moon was still well above the horizon and the sun was close to rising.

There was apparently little or no cloud, so unhindered moonlight would have provided a maximum of between 0.1 and 0.2 lux during the whole of the blackout, confirming Haas’s statement that there was enough illumination outdoors to walk around in. Building shadows in combination with glare from motor vehicle headlights and emergency lighting would have made moving about more difficult, but it was not completely dark. The illuminance in open shade may have been as low as 0.01 lux in cities and towns, and even less in rural areas.[27] It would be a mistake to equate even the dimmest of these conditions with darkness. These levels are still a few factors of ten above the human absolute visual threshold. Regardless of all this, the belief that darkness encourages anarchy and crime appears to have led to the statement “Luck, goodwill and a brilliant moon saved New York from disaster...” (attributed to Friedlander 1966).

The coincidence of the power failure, nighttime and reduced crime does not permit ready conclusions about direct and indirect effects. Lighting-related consequences might not be distinguishable from others having no connection with lighting, eg business closures because of inoperable electrical equipment, and disruption of electrically powered transport services. If police logs are still available for the night and the following daylight hours, it would nevertheless appear useful to compare them with others of the time.

3.2.5 New York City

Another large-scale power failure affected the whole of New York City and surrounding areas on 1977-07-03, starting at 8-37 pm Eastern Daylight (saving) Time (2037 EDT) and reaching the complete shutdown stage by 2136 EDT. Widespread looting and arson began within minutes of the shutdown. On the face of it, this seems to be invincible proof that darkness causes or at least encourages crime. Even in this extreme case, the facts show differently.

A whole issue of the magazine Newsweek (1997) was used to describe the social effects of the power failure on nine million people. The weather was oppressively hot and humid, in contrast with the cool mid-autumn conditions during the 1965 power failure discussed above. As the reporter put it:

“And in the gettoes and barrios of four of the city’s five boroughs, the looters and burners owned the night, on a scale and with a fury unmatched since the riots of a decade ago… But the switch-off of ’77 caught black New York in the midst of the summer’s worst heat wave and in the thrall of depression-level unemployment- and when the lights went out this time, the mean streets simply erupted. The arrest count exploded to a staggering 3,776 before the police gave up trying to collar the pillagers and concentrated on containing them.” Newsweek (1997, p 18)

About two thousand business places were looted and a thousand fires were lit. Despite many injuries and the theft of guns and ammunition, there were only two deaths, neither of them police (p 22). The looting and burning was unselective and total in some places but highly selective in others. One chain supermarket with allegedly high prices was emptied but a more popular supermarket nearby was untouched (p 26). “Not one mannequin was overturned on fashionable Fifth Avenue” (p 30).

Away from the gettoes, night life in the rest of New York went on just fine or even intensified. Cab rides cost four times the metered fare (p 21). Singles bars did near-record business (p 31), and a Chekhov play at the Lincoln Center finished under candlelight (p 21).

Corwin and Miles (1978) produced a 150-page study of the event, still available via a non-profit website devoted to archives and analysis of the 1965 and 1977 blackouts. Curiously, the document is still marked ‘Preliminary Copy’ on the title page but the content, much of it on the technical details of the power network failure, appears to have been done with thoroughness and insight.[28]

Corwin and Miles (p 15) gave the final arrest count as 2931 and the total number of arson and looting incidents as 1809. As temporary lighting was available at the Port Authority bus terminal, many people stayed the night there because of fear of crime in the dark (p 87). On p 91, Corwin and Miles stated “… power shortages of significant duration… provide the opportunity for some members of society to vent their anger at the social system.”

Corwin and Miles (p 104) listed 7 other major power failures in the US and Canada between 1971 and 1977, but these apparently were all of shorter duration and details of crime are not given. Appendix A to the report, ‘Documentation of societal impacts’, is of particular value. Using page numbers assigned here, interesting aspects are:

A8. Many more calls than usual were made for police service during the power failure, but the number of radio car runs was normal, despite the large force of police on duty. The additional calls were for information on what was happening or for assistance such as accident help, getting out of stopped elevators and arranging emergency transport to kidney dialysis.

A12. A ‘locational restriction’ in newspaper comments by authorities was a code telling the public that ‘minorities’ were involved.

A14. A Washington Post report stated that the public order in New York depended on where you were. Parts of the city had a holiday air, while others had the look of war.

A15. Comments from Boston built a picture of economic disaster, but from Washington the blackout was seen as a problem in social order. The Des Moines Register noted that most residents remained calm in the darkness.

A16. President Carter rejected a request to declare the city a disaster area.

A21. There was an increase in personal injuries from the looting melees and accidents in the dark.

A22. “On the whole, the blackout meant an increase in social activity, even in the riot areas.”

A23. A city health official reported an increase in the volume of ambulance runs due to stabbings, shootings and cuts from broken glass. But the actual data show a decrease in ambulance runs. “The police data above also do not indicate an increase in patrol car runs such as might be associated with crimes against persons.”

A25. “The initial reaction of the police and the courts was to respond to the looters as though the blackout was a disaster. The alternative, which in retrospect was closer to reality, was to view the looting behavior as that found during civil disturbances.” In contrast, disaster looting is generally committed by a small number of outsiders, who are sometimes from security forces sent in to prevent looting.

A26. “The police at first sent large forces to guard the major expensive shopping areas of downtown Manhattan and left the smaller local shopping areas with little- or no protection.” Looting during civil disorders is generally collective in nature: looters come from all sections of the community, often cooperate in pairs, groups, or even families, and loot selectively.

A27. The police response was often slow and ineffectual. An official report put the number of looters arrested as 3076, a conservative estimate of the number actually looting. “On any measure of looting behaviour during a disaster, this would represent an extremely large number of individuals, even for New York.”

The Executive Summary on p 2 states:

“The looting and arson that accompanied the blackout set aside the NYC experience from other similar power failures. While the blackout-related crime wave may have been a singular event, it nevertheless pointed out the delicate balance of our societal order, and the key role that electricity plays in maintaining that balance.”

Nowhere in Corwin and Miles (1978) is there any suggestion that crime was encouraged or increased specifically by the lack of artificial light, in contrast to the Newsweek quote that linked the start of the riots to “when the lights went out”. It was also when air conditioning and lifts (elevators) failed. Furthermore, there appears to have been an overall reduction in crime outside the areas of social disturbance during the 25 hours of the power failure.

Some additional information is available about natural lighting at the time. Full moon had been at 0324 UT on 1977-07-01 (BAA 1977), which is 2324 EDT on 1977-06-30, so the moon was almost 3 days past full[29] at 2136 EDT, the time of complete shutdown on 1977-07-03.

The following event times for New York City were obtained from USNO (2003).

For 1977-07-03:

Sunset 2030 EDT

Moonrise 2217 EDT

End of Civil Twilight 2103 EDT

End of Astronomical Twilight2235 EDT

For 1977-07-04:

Start of Astronomical Twilight 0325 EDT

Start of Civil Twilight 0457 EDT

Sunrise 0530 EDT

Moonset 0935 EDT

Sunset 2030 EDT

Moonrise 2252 EDT

End of Civil Twilight 2103 EDT

End of Astronomical Twilight2234 EDT

Shutdown was therefore 66 minutes after sunset and 33 minutes after the end of civil twilight. The moon rose 41 minutes after shutdown. There was still some faint trace of twilight left at moonrise because astronomical twilight still had 18 minutes to run.

An amateur astronomer (Rao 2002) took the opportunity of a lifetime to spend the night observing stars and galaxies with his telescope from The Bronx. He saw the Great Nebula in Andromeda and stars down to magnitude 5.4 with unaided vision. These observations are mutually consistent, and correspond to a moonless sky with about 2500 stars visible unaided and traces of the Milky Way overhead (Moore 2001). As it happens, this implies skyglow of about 1.6 mcd/m², a little fainter than the value shown (coincidentally) for 1977 in Figure 3. If the sky were uniformly this bright, the minimum illuminance at the ground would have been about 5 millilux (0.005 lux), or roughly one-fiftieth of maximum moonlight. This is consistent with the difficulty that people experienced in moving about New York on the evening of the blackout, before moonrise. This rather dim period lasted for less than an hour. Ambient light from road vehicle lights, emergency generator lights, flashlights, candles (about 1 lux at 1 metre distance) and even the flames from the gettoes would have supplemented this substantially in parts. After the moon rose, moonlight would have added about 0.1 lux at normal incidence or at most, 0.05 lux horizontal illuminance (Hollan 2003). This amount of direct moonlight is sufficient to walk about in. A slow further increase in illumination would have started at the beginning of morning astronomical twilight, about two hours before sunrise.

One indication of indifference to the amount of ambient light is that the looting and burning continued unabated for 25 hours, ie throughout the moonlit later part of the night and the whole of the daylight hours of the next day (Newsweek 1997, pp 22, 25). When it stopped next night, it was around the time that:

a. goods for looting and things for smashing ran out in the areas affected,

b. the power came on again, and

c. the last trace of twilight disappeared at the end of astronomical twilight.

It was also before the moon rose. At present there is insufficient information to decide which of these, if any, stopped the riots. (Doubtless, the police would have prevailed eventually, if not at this time.)

As a sophism, if an illuminance in the order of 0.005 lux is supposed to have triggered the race riots in this case, and perhaps also to have stopped them but only after they had continued in 0.1 lux moonlight and then in daylight, then it can be claimed that an illuminance of no more than 0.2 lux in the 1965 New York City case was enough to prevent race riots. But when the power cables failed in Auckland, the moon had passed last quarter and did not rise until after midnight, so 0.005 lux must have been enough to have stopped any race riots there before they even started! The argument is nonsense. What is true is that there was no increase of non-riot crime in these three cases, and in several others already mentioned, when artificial illumination was reduced or completely off at night. In the better-documented cases, crime actually diminished. The facts simply do not support Fennelly’s (1996) claim and Pease’s (1999) repetition of it.

At least one further New York City case of power failure is known but the details available are sketchy. During a heat wave that resulted in a heavy demand for electricity, 680 000 utility customers were without electric power on 1999-07-07. The failure affected business and residential customers, including about 250 blocks in the northern fifth of Manhattan. New York City police reported only a handful of attempted store break-ins and nine arrests (CNN 1999).

3.2.6 Amherst, New York

A ‘league table’ of crime in US cities and metropolitan areas of more than 75 000 population has been published annually for about a decade by Morgan Quitno, a Kansas company that researches crime statistics. The crime measure used is based on FBI annual Uniform Crime Reports (UCR) figures for murder, rape, robbery, aggravated assault, burglary and motor vehicle theft. Crimes in these categories are given a weighting according to surveys of what crimes pose most threat to people. For example, 66% of 501 respondents called burglary a serious or somewhat serious threat to themselves and their families, followed by car theft (61%), robbery (60.4%), aggravated assault (50%), rape (48.5%) and murder (40%). For each city, the sum of the weighted values provides a score that is used to determine the city’s place in the rank order (eg Morgan Quitno 1996). The resulting rankings tend to be relatively stable from year to year.

Amherst, a suburb of Buffalo in New York State, has consistently been at or near the safe end of the list of hundreds of cities for many years (eg Amherst PD 2002, Morgan Quitno 2002). Its low crime rate has been attributed to its suburban setting and affluent, well-educated population. Along with the other cities in the safest ten, its police force is at least 25% smaller than the national average, presumably reflecting the reduced need for police rather than the small force being the cause.

A pioneering outdoor lighting ordinance, initially against searchlights, was introduced in 1958 in Flagstaff, Arizona. Just a year later, Amherst, NY introduced stringent laws against obtrusive lighting, the relevant parts of which are well worth quoting verbatim here from Chapter 132, Lighting, of the Amherst Town Code (Amherst 2002):

“§ 132-1. General restrictions. [Amended 6-17-74, effective 6-29-74]

No person, corporation, partnership or association of persons shall use or operate or permit to be used or operated on their premises artificial lighting or illumination which unreasonably disturbs the sleep, comfort and repose of other residents or inhabitants in the vicinity, either by reason of the power or intensity or the location of said artificial lighting or illumination.

§ 132-2. Restrictions on lighting during certain hours.

"The illumination and lighting of such premises between the hours of 10:00 p.m. and 7:00 a.m. with more lighting and illumination than is reasonably necessary and proper for the safeguarding and protection of persons and property upon such premises shall be presumed to be lighting and illumination of such power and intensity as to annoy or disturb unreasonably the sleep, comfort and repose of the persons residing in the vicinity and shall be presumptive evidence of a violation of this section.

[§ 132-3 is about unauthorised road signs and lights.]

§ 132-4. Penalties for offenses. [Amended 5-31-66]

Every violation by any person, firm, association or corporation of any of the provisions of this chapter shall be an offense against this chapter and, upon conviction thereof, such person, firm, association or corporation shall be subject to a fine of not more than one hundred dollars ($100.) or imprisonment not to exceed three (3) months, or both, for each and every offense."

Amherst’s low crime rate is consistent with the situation described for other minimally lit places described above.

3.3 Small-Scale Darkness

It is clear from Section 3.2 that the unexpected reduction or failure of outdoor artificial lighting at night on different occasions and different locations has resulted in substantially reduced crime on geographical scales far too great to dismiss as some sort of local or chance aberration. Section 3.1 indicates that the crime suppression effect of dim and dark conditions operates down to school campus areas. An attempt is made here to ascertain whether the effect applies at even smaller scales.

In general, crimes are obviously more difficult to commit in darkness. In the case of graffiti and burglary, improving the ability to see with the use of hand-held artificial light sources occupies one hand or requires an accomplice and movement of the beam tends to attract attention. Regardless, it is a common belief that dark alleys, sideways, parks etc. in an otherwise brightly lit area are potential hiding places and crime locations for lawbreakers.

There are several ‘dark sky’ Internet list forums in existence. Most, if not all of them, have occasionally included anecdotes about thefts and vandalism taking place in brightly lit areas, eg cars parked under streetlights being singled out while others in darker places nearby are left untouched (eg Mizon 2002, p 197). Unsolicited anecdotes along similar lines have been heard by the writer and colleagues. Of course, these anecdotes could be dismissed as just another form of the juxtaposed extremes expressed in ‘sinning priest’ stories and the like, but many of them have been first-hand accounts, sometimes accompanied by photographs. Statements on the list forums are also made that street and school gangs have long met under street or other bright lights rather than in dark places (eg Hollingsworth 1995, Wilson 1998, p 8). Again the writer has heard similar unsolicited accounts.

The following, verbatim from one of these forums, presents a very different picture in comparison with the claims of some Situational Crime Prevention people:

“As a kid I lived in a smallish village midway between Liverpool and Manchester. There were only two streetlights one at each end. It was quiet, crime was non-existant, you did not have to worry about anything. Then the streetlights appeared. The fact that I had to have double curtains to be able to get to sleep at night was bad enough but the noise due to hooliganism and vandalism made it far worse. Vandalism was rife after the lights and IMHO this was a backward step and not the ‘improvement’ that had been intended by installing the lights. You could look out of your window and see the youths causing trouble. Before the lights they were just not there to cause any trouble… I now live in a village that has no streetlights at all and also no crime either. I do not think this is coincidence. Trouble makers and criminals need light just as much (possibly more so) than anyone else to see what they are doing at night.” (Eaves 2001)

Conventional lay views often appear to put undue emphasis on dimness or darkness as facilitating crime, possibly reflecting personal experiences of genuine fear in dim conditions.[30]

Most people do not have a good understanding of the situation-specific determinants of visibility such as the relative and absolute light levels, glare and ambient stray light. Adding more light would usually be the only solution considered for relatively dark areas where there is visibility loss from glare, but dim conditions are seldom mentioned in criminals’ assessments of where and when to commit crimes (eg Part 1, Section 3.2). Putting in more lights is too often just a quick and probably counterproductive fix that rewards the lighting industry for the poor quality lighting causing the problem. It may just displace or extend the difficulties. This chapter has provided the evidence to justify a new solution: get rid of the glare sources and unnecessary lights, all of them. Systematically dealing with light as a facilitator of crime at the small scale will go a long way towards fixing the large-scale problem.

4. MAKING SENSE OF THE EVIDENCE

4.1 Academic and Professional Responsibilities

In the preface to Pease (1999), the then President of the Institution of Lighting Engineers endorsed the review and stated that the (UK) “Home Office now supports the general consensus [sic] that well designed lighting, which is tailored carefully to the context of a particular area, has the potential to make a positive impact on crime”.[31] The Foreword, by a Chief Constable, includes the statements “Dr Kate Painter, and others, have, to some extent, ploughed a lonely furrow and often met with a wall of, at best apathy and at worst obstruction, in their valiant efforts in this area”, and “[I am] delighted that it [the report] bears out what so many people already thought.” The ILE published the review in accordance with a prior agreement to do so as part of an industry funding arrangement. The separate summary document (Pease 1998) does not mention this prior agreement, which was apparently an attempt to overcome possible objections about funding bias. A problem with such agreements is that they may take precedence over issues of scientific and technical quality.

Two other ILE publications also invite comment. The first of these (Painter 1999) is called ‘a guide for crime and disorder reduction through a public lighting strategy’. Apart from a title page style and content that is unusual for a technical document, it has six pages of text dealing with its author’s own work. The bibliography lists six works, five of which are Painter’s and the remaining one is Pease (1999).

The other publication is a Technical Report (ILE 1999), a guide for public authorities on the development of public lighting policy. It includes numerous mentions of lighting as a crime prevention measure. It encourages growth in the lighting and relighting of public places. It says nothing about curbing unnecessary and wasteful lighting installations and practices in the interests of greenhouse gas reduction and compliance with the Kyoto Protocol. Some of its photographs of outdoor lit scenes and objects appear to be endorsements of environmentally degrading practices including upwardly directed floodlighting and use of luminaires that send 50% or more of their total light output above the horizontal. It claims that a “large percentage of Skyglow is caused by light emitted directly upward or at high angles of elevation from poorly designed luminaires”. In fact, light travelling near the horizontal typically contributes about three times as much to artificial skyglow as a similar amount of light travelling upward at high angles. This is because low-elevation paths traverse long distances through lower levels of the atmosphere where the density of gas molecules is greatest and the concentration of light-scattering contaminants tends to be greatest.[32]

Fisher (1997) summarised the conclusions of the Tien, O’Donnell, Barnet, Mirchandani and Pitu (1977) lighting and crime study. He described a 1991 survey of fear of crime after relighting of an area in the UK: the results were beneficial but the effect on actual crime was not ascertained in the study. He accepted Painter’s work as establishing lighting as a crime prevention measure without displacement. He wrote of the “large and growing body of data which suggest that lighting can be an effective crime countermeasure” and suggested that the Australian Standard set for road lighting (SA 1999) should undergo a major revision “based on criteria which include the incidence of crime”.

Criticism of Painter’s methodology by Eck (1997) and Lab (1997) cast doubt on lighting as a crime countermeasure. Studies published within the next year or so did not alter the balance of evidence against lighting for crime prevention. Despite this, the joint Australian /New Zealand Standard, AS/NZS 1158.3.1: 1999 Pedestrian Lighting (Category P), issued in 1999, frequently mentions lighting not only as a means for reducing the fear of crime but also as a means of preventing crime. Six references are listed in Appendix B of this standard: two are British Standards, two are publications of the International Commission on Illumination (CIE) dated 1992 and 1995, one is Fisher (1997) and the last is a publication of 1994 called Crime and lighting and attributed to ‘The Institution of Public Lighting Engineers’.

Standards Australia has since publicly advertised AS/NZS 1158.3.1: 1999 extensively with emphasis on the need for public authorities to purchase and follow this standard in order to help prevent crime. As in the UK, public authorities have doubtless thereby been influenced to install more and brighter outdoor lighting. Consequential unnecessarily incurred expenditure could become an issue.

The International Commission on Illumination (CIE) has issued a guide to the lighting of urban areas (CIE 2000). Its recommendations cover the supposed effect of lighting on crime at night. Like the standards and guides just mentioned and many others, it now needs substantial corrections. The Institution of Lighting Engineers has issued guidelines for reducing light pollution (ILE 2000), based on CIE publications. It specifies zero upward light ratio (ULR) for luminaires used in intrinsically dark areas and ULRs increasing to 15% for luminaires used in town and city centres. If 15% seems a lot of light to pay for and then send into outer space, consider the maximum upward waste light ratio (UWLR) allowed for pedestrian lighting in AS/NZS 1158.3.1:1999, an extraordinary 40%. Worse, many local councils, companies and individuals in Australia continue to ignore Australian Standards in installing and maintaining fence-top, pole-top and cluster-mount globe luminaires that send as much as 60 % of the emitted light above the horizontal when the globes are clean and 70% or more when they have collected grime as usual on their lower parts.[33]

Erroneous scientific hypotheses and flawed investigations contribute to human progress insofar as they stimulate debate and further investigation. Only rarely are scientific papers in scientific journals withdrawn and even then they remain physically available in distributed copies of the journals. But continued unqualified propagation or citation of the material after it has been disproved or discredited retards progress. For example, Grabosky and James (1995) includes an uncritical two-page summary of Painter’s London project, headed ‘Crime prevention and fear reduction through enhanced street lighting’. It also includes a summary of ‘Public transport safety in Victoria’, which involved multiple treatments including overbright lighting for railway stations.[34] The Grabosky and James document, sub-titled ‘Leading crime prevention programs’ was reprinted in 1997 and was still available online, apparently unchanged, into 2003. Many other crime prevention articles and books from apparently reputable sources likewise continue to promote the supposed crime prevention effect of lighting, regardless of the existing evidence against any such effect. Academic and professional bodies involved might usefully reconsider their scientific publication standards.

Neighbourhood Watch (NHW) and similar organisations are well established in Australia and elsewhere. NHW publications frequently state or imply that lighting is a crime prevention method,[35] encourage the installation and use of outdoor domestic lighting to prevent crime, and encourage immediate complaints to local authorities when streetlights fail and supposedly increase the risk of crime. Advice by the writer to various branches of NHW about their circulation of such erroneous advice has frequently been met with disbelief and even ridicule. Subsequent propagation of the established view seems to have continued unabated and without qualification.

Finally, for all the value of the Internet, it has the downside of allowing unprecedentedly rapid dissemination and reinforcement of urban myths. The Council of Europe (2000) encouraged local authorities to light up all their public places “to give citizens a sense of safety”, despite the gross levels of outdoor lighting and lighting waste that already beset much of Europe. The Internet is figuratively awash with encouragements for anybody and everybody to light up campuses (eg Sowell 2001), buildings and outdoor areas in general as a supposed crime prevention measure,[36] usually with no attempt at justification. Many of these ill-informed statements are in ‘security’ websites hosted by government departments, local government authorities, police forces, community organisations and service clubs. Often there are assertions such as ‘It is well known’ or ‘Research has shown’ that ‘lighting will prevent crime’ (not even just ‘deter some crime’). Some of these messages are reproduced on dozens of websites, often without acknowledgement of the source but copying or adding spelling and grammatical errors.

Lighting associations and professionals in general should ensure that information provided is factual unless qualified as something else. Otherwise, eventual users of the information may commit to expenditure that is unnecessary or even wasteful.[37]

4.2 What is the Form of the Lighting and Crime Relationship?

4.2.1 Direct and indirect contributions to the total effect

As shown in Figures 1 to 3 and Chapter 3, there are positive correlations between the crime and lighting time series studied. But theFarrington and Welsh (2002a,b) meta-analysis result (see Part 1) indicates the opposite for data from the UK and the USA. It seems inescapable that the meta-analysis does not provide credible guidance about about any real overall effect of lighting changes on crime. At least at the scales considered in this document, the true odds ratio must be less than 1. This is on the assumption that the time constants for crime response to lighting changes are not in the order of say a decade or more. Such long delays in response appear unlikely, given the observed quick reduction in crime following a city blackout. Even if the true overall odds ratio is just less than 1, this is substantially less than 1.18, the lower value of the 95% confidence interval given by Farrington and Welsh.

Marchant (2003) pointed out that the quantity called the odds ratio by Farrington and Welsh would be more correctly described as a cross-product ratio (CPR). He also showed that incorrect statistical treatment by Farrington and Welsh resulted in values of the CPR that were too large and confidence intervals that were too narrow. As with the errors described in Part 1, the net effect is to increase the likelihood that the true value of the CPR is less than 1. Such a value would mean that crime increases with the amount of outdoor light, consistent with the observations described above.[38]

The idea that outdoor crime at night has a net positive direct connection with outdoor light levels may seem counter-intuitive, given the common experience that lighting tends to allay the fear of crime. If true, it would imply a predominance of daytime crime over night crime, as actually happens with residential burglaries in the USA. Research would be required to establish the relative contributions of light and social factors in determining such predominances, but some general observations are possible here.

If all crime did decrease steeply without temporal displacement as a direct effect of increased lighting, and if night and day opportunities for crime were equal, turning night into day with artificial light could be expected to do no more than halve the overall crime rate. But some crimes are generally committed indoors (eg ‘white-collar’ crime). Any direct facilitating influence of outdoor lighting at night on such crimes does not seem credible. Therefore, for all kinds of crime aggregated, the maximum possible decrease accompanying the installation of intense outdoor lighting, according to existing belief, would appear to be limited to less than a factor of two.

Ambient outdoor light levels at night in many parts of even the world’s most brightly lit cities are still well short of daylight values in general. The actual crime rate increases shown in Figures 1, 2, 3 and 4 generally well exceed a factor of two over two decades or so. A positive (adverse) direct effect of lighting on crime is in accordance with the long-term direction of change in observed crime but the national increases in crime rate appear too large to be accounted for by direct effects of lighting increases. The crime-changing effects of lighting claimed by Painter and Farrington (1997, 1999a,b) do include what are here called indirect effects and could account for the magnitude of the observed changes, except that they are clearly in the wrong direction. Any direct effects of lighting on crime appear to be swamped by other processes.

Excluding temporal displacement, a direct daytime effect on crime, increase or decrease, by what some commentators have called ‘switched-off outdoor lighting’ can hardly be regarded as anything but unlikely if not far-fetched. Nevertheless, there are known environmental and socio-economic effects on crime incidence, such as seasonal effects (Baumer and Wright[39] 1996, Jochelson 1997, DCPC 2001) or, say, the effects of wars or overseas financial changes on international tourist numbers and hence numbers of tourists available to be crime victims. These fit the above definition of indirect effects. Indirect social or economic effects of changed night lighting on daytime crime incidence seem perfectly reasonable to consider, as Painter and Farrington, Pease and others have done. But in the absence of firm knowledge or good reason about the direction of effects being investigated, it would seem important to keep an open mind about the individual directions for day and night in the course of analysis. It might be possible, for instance, to have an indirect effect in which the night and day segments had opposite signs or different magnitudes, or both. If day and night effects were opposed to any extent, however, it seems unlikely that the overall growth of outdoor artifical light would have been accompanied by the observed substantial growth in the overall crime rate.

The positive correlation of lighting and crime does not necessarily imply causality. The quantities may be causally unrelated but both dependent on some other factor(s) such as economic growth, or causally related directly or through some other processes, or dependent on a combination of these. Unexpected outdoor light reduction at night in cities is followed quickly by crime reduction, however. This an independent confirmation of a positive correlation between light and crime. Although it is not absolute proof of a causal effect of light on crime, it is strong evidence for causality.

4.2.2 Form of the variation

4.2.2.1 A graphical approach

Figure 6 encapsulates key facts and suppositions about ambient light and crime. The horizontal scale extends over most of the range of illuminance found outdoors from the darkest of overcast moonless natural nights to tropical noon daylight. Line A indicates the illuminance range within which lighting and crime experiments have usually been performed. Line B represents an increase of 3.375 times in illuminance, a typical treatment in experiments to date (see Part 1).

The assumption is made that crime will not be affected materially by variations in light level during the daytime. This allows the mean daytime crime rate for a particular place to be used as a benchmark for crime occurring at light levels lower than those encountered in daytime. All crime is plotted as a fraction or multiple of the benchmark. The crime levels plotted in the figure can represent quantitative observations for a particular place, means for a number of places or assumed characteristics in a theoretical development.

Some fraction of all crime observed at a particular non-daytime light level will be a response to social conditions including social activities tied to time of day, along with any non-lighting effects attributable to weather factors (eg Cohn 1993) such as extreme temperatures or rain or snow. The remaining fraction, which could be anywhere between zero and all crime at that light level, will be a response governed by, ie causally related to, the light level. This may unintentionally include any non-lighting effects closely coupled to light level, as with the regular monsoonal evening downpours experienced in some tropical cities. If not recognised and allowed for, any non-lighting-caused effects could reduce the reliability of predictions of crime changes expected from artificial lighting changes.

Usually, a graph showing crime by night and day will be drawn against clock hours. The mean light levels for any particular clock time will depend on geographical position, season and weather. The social conditions that determine crime (eg Glaeser and Sacerdote 1999, Ousey 2000) and non-lighting environmental conditions prevailing for any particular light level will be linked to clock time (including variations such as daylight saving time) and the seasons.

Social factors such as time of day outside of school hours (NCJRS 1999) and work hours are known to play a large part in the incidence of crime, as described in Part 1 and Section 1.2 above. Large fluctuations in the rate for many types of crime occur consistently at fixed clock times, usually without any definite or consistent link to specific light levels. To single out any specific effects of light changes, at sunset or sunrise for example, crime would be better plotted against phase of day and night rather than against clock time.

Lunar phase and meteorological factors such as cloud and fog can have a profound effect on the light level at a given day/night phase, however, so it would be better again to plot crime against measured light level. Nobody appears to have pursued this so far, which sits strangely against the vast amounts of government, corporate and domestic expenditure on artificial light at night as a supposed crime-prevention measure. As an example from highly developed countries, the contrast with requirements for evidence-based official approval of pharmaceuticals is extreme.

Even when daylight saving time is not in use, the use of clock time or even local solar time as a proxy for ambient light levels over night and day results in the integration of any crime effects of light level over light stimulus values varying in some cases from the assumed mean by ten times (1 log unit) or more in each direction. Unless the observations and analyses involved can be conducted in more than two dimensions, the cost of avoiding this reduction of apparent effect of light would be to integrate the other effects instead, over a substantial and frequently ill-defined range. In terms of the effects on a light and crime graph, this would effectively ‘smear out’ fixed-time-of-day crime effects along the light axis, thereby increasing the prospect of being able to identify crime effects attributable to absolute or relative light levels.

4.2.2.2 Possible forms of crime and lighting curve

Curve C in Figure 6 is representative of the form of variation of crime with ambient light implied by conventional Situational Crime Prevention theory. Its form, let alone its possible partition into light-dependent and other parts, appears to be ignored in the literature. The observational evidence supporting a curve of this form is that aggregate crime in some countries (eg the UK) is greater at night than by day. However, it is often overlooked that clock time rather than light level appears to be the main factor governing night or day prevalence in some cases.[40] As the ambient light is increased artificially or increases naturally from night conditions, separately or together through twilight values, by definition the crime rate indicated by the curve must reduce down to the daylight level, whatever that may be in absolute terms.

Although the Curve C shown is notional, the vertical scale shown is realistic to the extent that the net excess, integrated over the range of light levels experienced at night, is at most a small multiple of the day rate. In Figure 6, the excess over the day rate is arbitrarily set at a factor of 4 for 0.0001 lux, an illuminance that is near-blackness in terms of visual appearance. Curve C is monotonic (ie having a slope that does not change sign), the simplest plausible form of variation.

A problem with Curve C is that there appears to be no reliable evidence to back up the belief of at least some Situational Crime Prevention people in a tendency for rampant lawlessness and anarchy in near-darkness. In fact, as the ambient light becomes too dim for any visual response, the crime rate could be expected to level off as shown by the dashed alternative dim-light end for Curve C, or even to drop, instead of continuing to rise steeply. This logical need for an asymptotic or non-monotonic dark characteristic does not appear to have been considered by theorists to date.

The form of curve for fear of crime found by Boyce, Eklund, Hamilton and Bruno (2000) is rather like Curve C, having a high value in dim light and dropping down towards the daylight value as illuminance increases. The Boyce et al. results only extend down to about 0.1 lux so the flattening off that could also be expected for the fear of crime curve at very low light levels does not appear in their graph. Nevertheless, the form of Curve C illustrates the popular notion that actual crime increases with reducing light. This belief might be based on common experience that the fear of crime tends to increase with reducing light at night, together with failure to distinguish adequately between fear of crime and actual crime.

Curve C illustrates an important point discussed at length in Part 1. If this curve does truly represent the actual situation, it indicates that a typical lighting treatment of about 3.375 times increase in ambient light at night (or 0.53 log unit) would produce a rather small reduction in crime. For Curve C as drawn and the lighting increment B as shown, the reduction is about 9% in crime, ie from its initial value, x, to 0.91x. Crime reductions claimed in some experiments in the UK have been as much as this value compounded fourteen times.

As an example of an excessive claim, Farrington and Welsh (2002a,b) calculated the daytime indoor crime reduction was to 1/3.82, or 26%, in the Birmingham market hall relighting of Poyner and Webb (1997). Suppose that the market hall with its original lighting had twice the crime rate that it would have in full daylight. The crime reduction claimed for increasing the light by some unknown amount (Poyner and Webb did not give a numerical factor for the light increase) is so great that crime must have been reduced to about half of the daylight value, an illogical result. Another possibility is that the original crime rate was four times the daytime rate, and really intense relighting changed near darkness to daylight levels. However, the writer is not aware of any reports of such high rates being observed anywhere as an identifiable consequence of dim lighting.

Line D in Figure 6 represents the case where crime is completely independent of ambient light, and all non-lighting effects balance out. This is a possibility indicated by experiments in which the null hypothesis (viz, that light has no effect) is not disproved. It would also be indicated by a meta-analysis cross-product ratio that was not reliably different from zero. To date, typical lighting treatments have usually been such a small part of the daylight-to-darkness range that it would be a large extrapolation to claim light-independence over this whole range.

Curve E is the simplest form of variation that is consistent with the low crime values observed in outdoor lighting blackouts (or, more accurately in terms of residual light levels, as ‘dimouts’). By itself, outdoor darkness could hardly be expected to stop crime such as fraud, Internet offences, dangerous driving, or domestic violence, so the overall rate would not drop to zero. In the absence of any evidence to the contrary, the ‘true’ curve must nevertheless start low down at the dark end of the illuminance axis. Then it must rise monotonically to asymptote to the daylight crime level, like Curve E.

4.2.2.3 The general form of lighting and crime curve

If there is a true form of the variation of crime with lighting, then it should be consistent with all observations. According to Curve E, crime increases with increased light up to daylight levels of crime and light. If light levels and not social circumstances determine crime levels at night, then it could be expected that the numbers of crimes at night would always be less than the numbers of crimes by day. Because education, work and community service activities are predominantly in daylight hours, however, it could be expected that there would be more potential criminals at leisure and available to commit crime in the evening. Likewise, there may be more potential victims out and about after the end of school hours and work hours for the day. Reports of the proportions of crime by day and night generally do not include details of such factors. In this case, the known social factor of leisure time is expected to act in the direction of increasing the percentage of crime that occurs at night.

As mentioned in Section 1.2 above, some reports show a night excess of crime, and others, a daytime excess. Figure 5 shows both: for many years in the USA, there has been a greater percentage of residential burglary by day than at night, while it is the other way around for non-residential burglary. This is clear evidence against a single ‘true form’. ‘General form’ or ‘net form’ would be better, especially if crimes aggregated across types have to be considered in preliminary work. The net result from Figure 5 is that a greater percentage of all burglaries take place in daylight, so a general curve for burglaries would have to show a net reduction of burglaries at night in comparison with burglaries by day.

The actual number of burglaries in the USA has been declining for many years. This does not invalidate the observation (Figure 3) that outdoor artificial light and overall crime are growing together, especially given that increased police numbers and increased incarceration have been acting to reduce crime, as mentioned in Sections 2.3.3 and 2.4.

Because there are substantial fluctuations in the crime rate observed wholly within bright or dim light conditions, day or night, some unknown part of any general changes in crime indicated by Figure 6 will be a result of whatever social factors apply in particular cases. The available evidence supports the presence of substantial effects both from light and social factors. Because of the time delays associated with some of the social effects (which would include the social changes responsible for the indirect effects of light), all of the possible forms of curve in Figure 6 would be best regarded as representing equilibrium states rather than short-term effects.

The available evidence certainly appears to favour Curve E over Curve C or Line D, especially at low light levels. The case is far from conclusive for higher light levels, however. The general form might be a mixture of any two of these forms of variation, or even all three, depending on the mix of crime, social influences, and the balance of beneficial and adverse effects of light, direct and indirect, in aiding and hindering crime. Instead of a single curve, the reality could be an area between or covering the curves, perhaps wider in some parts than in others.

Another possibility is that the general curve might not be monotonic. It would still need to start low down in accordance with available evidence, but may cross the axis and asymptote down to the daylight crime rate at higher light levels much as Curve C does. This is Curve F in Figure 6. Increasing light at night would therefore have an undesirable effect on crime up to the light level corresponding to the peak of this Curve F. At and near the peak, small lighting changes would have little effect. At higher light levels along Curve F, increased lighting would have a small beneficial effect on crime, ie a reduction. Overall, however, there would still be a net adverse effect of increasing illumination artificially above natural night levels. An attraction of Curve F is that it might help to explain the ‘mixed’ results of some lighting and crime quasi-experiments to date, instead of supposing that they are entirely a result of confounding by unknown real-world factors.

A further possibility that could explain mixed results is based on Curve E. This form would not be a single line but a band representing the crime variation that undoubtedly occurs in the real world at specific light levels. This band could be thought of as representing the range of variation within statistical confidence limits. At this stage, not much can be said about the likely vertical extent of what could conveniently be called Band E (not shown in Figure 6 to avoid excessive detail), except that it may be bounded on its lower side by zero crime, at least at the dim light end, implying a Poissonian distribution. The upper part of Band E may well extend above the daylight crime line at any light level but would seem more likely to do so as it approaches daytime light levels.

At daytime light levels, Band E is constrained to coincide with the level of variation applying to crime in daylight. In this diagram, the band will have a much smaller vertical extent because it is integrated over daytime hours as a consequence of the definition of the daytime baseline. Band E, low down and broad at low light levels, therefore converges by definition to a thin horizontal band at high light levels. At below-daylight levels, Band E could readily extend above the horizontal line marking the daylight level, accounting for the existence of elevated levels of some crimes in some places at night. The mean or perhaps the geometric or logarithmic mean of the vertical extent of Band E would be marked by Curve E.

4.2.2.4 Linking theory and observation

The daylight end of Curve E is fixed. The low light end is in general accordance with observations of low crime rates in near-dark conditions. If the direct effect of light in increasing crime were the sole mechanism acting, the whole curve would be fixed. However, the indirect effect of light is not only present but appears to predominate, at least over the illuminance range indicated by Line A. Its effect is to change crime through some social action, generally with an associated time delay. The present meagre evidence indicates that the net indirect effect of increasing light at night is to increase crime by day and night more or less equally. A change in light at night would therefore result in some change in the absolute level of crime by day, in due course.

In deriving the overall result of a lighting change on crime (indirect plus direct effects), account would need to be taken of the relative change in crime at night over the duration of night, plus the relative change over daylight hours. The transition between day and night might need to be dealt with as some intermediate change over twilight hours. The net effect would be an average of crime weighted by duration, taking account also of the absolute crime rate represented by 100 on the vertical axis. With this information, eventually it should be possible to construct graphs showing the relationship between the overall crime rate and the mean light level at night in individual places of any size. Conversely, it may be possible to work backwards from observational data of overall crime rate and light levels to derive the general form of curve applying to the axis variables of Figure 6, bearing in mind the likely presence of ‘smeared-out’ non-lighting social effects related to clock time.

4.2.2.5 Practical issues

So far, this discussion has related to light and crime at an individual place. But even with a single streetlight and its surrounds as ‘the place’, the actual illuminance or luminance will vary greatly with geometry in the vicinity. This raises the issue of what single light-related value on the horizontal axis is to be used to represent the actual situation at night on a graph of Curve E. The luminous flux emitted by the luminaire and related photometric quantities such as peak illuminance take no account of the area lit or number of individuals served. The use of quantities such as lumens per unit of ground area or light power per person would appear to meet the immediate need for some sort of average value as a single number. The place or area in consideration could equally well be under a streetlight, a town, city, groups of cities or the whole of a country. The practical upper limit for size might be set not by the light-related measure but instead by the homogeneity of the criminal justice system or systems involved. Even then, some sort of continent, hemisphere or global figure might be useful as a benchmark in research and planning.

The quasi-experimental work reviewed in Part 1 might now be seen to have previously unrecognised faults, in that little or no attention was directed to features of Curve E that need to be taken into account in the design, conduct and interpretation of such experiments. Future experimental designs will need to incorporate sufficient sensitivity for reliable detection of smaller changes in crime than those found to date. Treatments much larger than hitherto used could assist, but run the risk of a confusing result if the actual effect varies rapidly with light level or is non-monotonic. And, as emphasised in Part 1, much more effort needs to be devoted to measuring the actual light levels at representative places in the experimental, control and adjacent areas throughout the whole period of the experiment. In cases like Painter’s Dudley and Stoke-on-Trent experiments, where the before and after periods were 1 year, for example, it will not be good enough to start the experiment at the before interview period, just before the relighting treatment is applied. Monitoring of crime and light levels must be in place for the whole of the before period and the whole of the after period, preferably with the aim of plotting crime against actual light level as well as against light phase, clock time or integrated time.

4.3 Spatial and Temporal Relationships

For all US cities with a population of more than 1 million, the rate of violent crime in the central city areas is between 1.22 (Las Vegas, NV) and 7.63 (Milwaukee, WI) times that for the respective suburbs, with an average of 3.22 (Demographia 1999). The Total Crime Index rate for US rural areas is less than half that of metropolitan areas (Maguire and Pastore 2002, Table 3.121). The situation is similar in other countries (Walker 2002).

Common experience is that cities are more brightly lit than their suburbs, which in turn are more brightly lit than rural areas. The same applies to towns, their outskirts and rural areas. Air travellers at night are often able to see these progressions at a glance.[41] The progressions are consistent with the diminution of artificial skyglow as distance from cities and towns increases. From the known experiences of urban blackouts and reduced crime, the important part of the progression is not one of infrastructure, big buildings to detached houses to isolated farmhouses, but from lots of artificial light at night to little or none. This widely evident spatial relationship suggests a positive correlation between the amount of artificial light and the crime rate. This does not imply causality, although it is additional evidence to support causality. The two quantities could be mutually independent but both dependent on population density, for example. Further evidence is needed.

As shown in Table 1, crime rates and notional skyglow growth are positively correlated over time in each of five countries and similar temporal correlations are likely in 15 others according to Figure 4. The crime data sets cover national populations and durations of decades to a century. The skyglow growth curves are based on observations and cover a century on the basis of reasonable assumptions. Their exponential form is consistent with observations from other countries and satellites as an indicator of the accelerating growth of outdoor ambient lighting in most places.

Street lighting reductions in certain US towns have not resulted in reported increases in crime or road traffic accidents. Casual observations indicate that graffiti and probably other forms of vandalism are deterred by darkness. Anecdotal reports of Dark Campus programs indicate likewise, along with other crime reductions. Power disruptions in cities have likewise resulted in dim conditions and reduced crime. Dim lighting or darkness therefore inhibits the crime rate both as a spatial and temporal effect. None of this supports the notion that insufficient light at night increases crime. Light and crime are positively correlated in the circumstances described,[42] whatever people might say or think.

Several quasi-experiments that purport to show differently appear on examination to have too many shortcomings for their results to be accepted as reliable. These and other experiments are suspect for various reasons including uncertainty of treatment magnitude, unduly large effects, financial and non-financial conflicts of interest, confounding, and hotspots regressing to the mean independently of treatment. The collective total data for these experiments is minuscule by comparison with the range and magnitude of the data used so far in this paper to reveal spatial and temporal associations between lighting and crime, viz:

· populations of hundreds to hundreds of millions,

· crime data from small areas such as streets and campuses up to national scales,

· spatial extent also ranging from small areas to national scales,

· dark or dim durations from a single night to thousands of nights,

· historical data extending to over a century, and

· lighting industry literature and standards, published and ad hoc outdoor photometry, appearance of outdoor lighting at night, anecdotal evidence and measures of visibility of celestial objects, skyglow measures and photographs, and satellite-derived measures of upward light emissions.

A negative correlation between lighting and crime, long claimed by Situational Crime Prevention practitioners and others, appears to be an unwarranted intuitive extension of the lighting and fear of crime relationship to actual crime. No reliable evidence for this negative general relationship has been found. ‘Dose-response’ curves are known from experiments about fear of crime. Fear tends to diminish with increasing light, so the curves typically have negative gradients (eg Boyce et al. 2000, Fig. 7). Specific curves for lighting and actual crime are unknown but typical net effects tend to have positive gradients. Whatever residual experimental evidence there might be for a non-zero inhibiting effect of lighting on crime, it appears to be overwhelmed by the collective evidence for the opposite effect. [43]

The issue of whether lighting is an active causal factor, or is merely a co-variant of something else such as population density, may usefully be studied within a theoretical framework. This framework is now developed around the evidence already established, mindful of the difficulties of working with field data that are much more open to confounding than data collected in controlled experiments. In addition, while correlation is a necessary but not sufficient condition for making confident inferences about causality, it does not matter which statistical technique has been used to establish correlation (eg Wuensch 2001).

4.4 Lighting and Crime Hypothesis

4.4.1 Possible processes and causality

Proponents of lighting for crime prevention assume or believe that lighting at night will reduce crime. Published results of quasi-experiments sometimes indicate this effect, sometimes no effect and occasionally even an increase in crime. The question examined has usually been ‘Does lighting reliably reduce crime or have no effect?’ Clearly a more general form would be ‘Does lighting reliably affect crime or not, and if it does, in which direction is the effect and how big is it?’ But even this ignores at least three other possibilities. Firstly, the relationship between lighting and crime could be chaotic in the mathematical sense.[44] If so, it would be possible for conventional before-after testing at various times and places to produce statistically significant results in both directions of effect. Although a chaotic association is a possibility, it seems unlikely considering the consistently positive associations found in temporal and spatial data. It is not the simplest available explanation of the observed facts and is therefore disregarded here.

Secondly, the causal direction could conceivably be the other way: ‘Does crime reliably cause lighting or not, and if it does, is the effect positive or negative?’ Even though this might seem nonsensical at first sight, there is ample common experience to indicate that there is indeed an effect and that its net effect is positive. For example, consider a break-in at a house. The experience may motivate the victim to fit extra or better locks and to change the lighting arrangements. New or increased ‘security’ lighting would be a typical outcome, either all-night or movement-triggered. Thus, crime has led to, or ‘caused,’ lighting or more lighting as a reaction. At present, it would be rare for the response to be reduction or removal of lighting, so the overall effect is strongly positive in the case of burglary. But vandalism of outdoor lighting and parts of the electricity distribution system can and does directly cause darkness or at least dimness over areas ranging from a single streetlamp coverage to city and regional areas. Intentional disabling of lighting to hide a criminal act is also known, along with the use of light to facilitate crime.[45] Thus, crime does cause lighting in both the positive and negative senses. The net effect appears to be positive, in that lighting installed as a reaction to crime appears to exceed the amount of lighting disabled by vandalism, especially when durations of normal operation and disrupted operation are included.

The third possibility is a combination of this ‘crime affecting lighting’ effect with its more conventional inverse, ‘lighting affecting crime.’ Observations support the proposition that the relationship is bi-directionally causal. This relationship is included in the following description of a hypothetical often-cyclic process that necessarily involves commerce.

4.4.2 Lighting, commerce and crime processes – a new hypothesis

Suppose that an urban centre of any scale has an incidence of crime sufficient to cause public disquiet. Especially as elections approach, politicians promise action such as more police, homeless youth outreach and so on. Often the list will include ‘improved’ outdoor lighting, one of the less expensive and more obvious ways for politicians to be seen to be doing something about crime. Increased lighting is provided in due course, probably both as a reaction to actual crime and as a pro-active attempt to reduce future crime. At night, it tends to reduce fear of crime and attracts more people, which may sometimes result in an extension of shopping and business hours in the area at night. In any case, commerce may improve sufficiently to allow increased investment in goods stocked, services and facilities, and in infrastructure including increased outdoor lighting and advertising signs to attract more customers. In turn, the investment encourages additional utilisation of the commercial area in the daytime as well as at night. The improved prosperity increases job opportunities, which increases pressure on housing in surrounding residential areas. Housing and population may increase in both numbers and density.

Crime tends to increase as a reaction to the presence of more people, more money and more goods[46] brought about indirectly by increased lighting. From experience and experimental evidence, any net direct effects of increased outdor lighting in aiding or hindering the commission of crime are generally small and necessarily confined almost completely to nighttime by definition. The overall effect of increased outdoor lighting at night, largely via commerce, therefore tends to be an increase in the crime rate forday and night together. This is consistent with national experiences over many decades (eg Figures 1 to 4).

The process generally appears to be cyclic, not necessarily in a regular pattern. Some outdoor lighting is installed primarily because of, or ‘caused by’, crime. It may be an attempt to prevent crime, or a reaction to crime that has already occurred.[47] Increased outdoor ambient light tends to allay the fear of crime and thereby to increase commerce, which indirectly leads to or ‘causes’ more crime. Two-way causal connections between outdoor lighting and crime appear to apply just about everywhere, almost regardless of scale, not just in developing urban centres. Naturally it could be expected to be a much slower process in the case of adding a light or two outside an isolated country store.

Casual observation indicates that the installation or increase of lighting is more commonly~~pro-~~ pro-active than reactive: crime more frequently appears to be a result of increased lighting than increased lighting is a result of crime. For example, generally the lights are operating before a new shopping mall is opened for business. The amount of lighting installed primarily as a reaction to crime appears to be a minor part of the total of all outdoor lighting and is unlikely to account fully for the observed strong spatial and temporal correlations. For night and day combined, therefore, the direct plus indirect effects of all outdoor lighting generally increase crime more than they reduce crime. As mentioned, this is an outcome of a bi-directionally causal relationship.

In accordance with Curve E of Figure 6, the crime-increasing effect of lighting increments would presumably asymptote to zero as night ambient artificial light levels approach those of daylight. The total crime experienced would then be determined largely or only by the applicable social factors. This raises the possibility of using daylight lighting levels at night to separate out the social factors. Given the likelihood that crime would increase as a result, ethical considerations may block such experiments.

In commercial areas, a substantial proportion of all outdoor lighting, including lit signs, is installed primarily to increase commerce by attracting customers. To the extent that the rest of all outdoor lighting generally encourages people to be out and about at night, it also assists commerce. Commercial prosperity leads to company expenditure on promotional lighting, assists in the provision of public lighting through taxes, and is presumably a factor in the affordability of domestic outdoor lighting for employees and in the number of employees who might consider installing such lighting. Thus there is a bi-directionally causal relationship between commerce and lighting. A similar relationship seems likely for commerce and crime: increased commerce would appear to motivate criminals and increase the opportunities for crime, while crime would appear to inhibit commerce. Lighting, commerce and crime therefore appear to be related causally in most or all of the six possible directions between pairs. On the evidence presented in this document, there is observational support for the notion that lighting is a cause of crime, regardless of whether this comes about largely through intermediate interactions with commerce or some other mechanisms.

As a means of increasing commerce, outdoor lighting also seems to be an important contributory factor in the processes of urbanisation, urban intensification, urban utilisation and urban sprawl, thereby possibly further accentuating growth in the crime rate. This is not in conflict with the existing view that urbanisation “in any country generally begins when large-scale commerce takes root and most new jobs are to be found in the factories and financial centers in cities” (NASA 2000).

The six possible interactions are indicated in Table 2, along with effects and reasons that appear to be important. The sign of an effect within an interaction is given as positive when an increase in the causative variable produces an increase in the dependent variable. Conversely, a negative sign indicates that the affected variable would decrease. Apparently strong effects are indicated notionally in Table 2 by two or more signs. Two signs do not necessarily imply twice the effect of one sign, but just a heuristic estimate that the effect is generally stronger. In due course it may be possible to quantify the strength of interactions in particular examples, eg as partial correlation coefficients and their positive and negative constituents.

The entry marked * in the ‘Reasons’ column of Table 2 is the converse of imposed darkness inhibiting crime, which is clearly causal from observations. It is strong evidence against the notion that crime and lighting are unconnected quantities merely growing concurrently because of separate reliance on economic conditions.

Here is a concise statement of the new hypothesis:

Outdoor artificial light, commerce and crime are causally interactive in all six possible directions. The sign of effects in each interaction can be positive or negative, or a mixture of both. Light leads commerce and commerce leads crime as dominant overall effects. Light may lead or lag crime. The net overall effect is for light to lead crime.

TABLE 2. Apparent Strengths and Signs of Hypothesised Effects
Causal Variable		Affected Variable	Notional Strength and Sign of Effects	Reasons
Lighting	causes	Commerce	+++ -	Pro-active attraction of customers Lighting costs reduce profits
Commerce	causes	Lighting	+	Profitability allows increases
Lighting	causes	Crime	+ + + + +, -	Lighting decreases fear, increases risk Lighting increases urbanisation Lost inhibiting effect of darkness* Facilitating graffiti Direct aiding and hindering by light
Crime	causes	Lighting	+ + -	Lights installed for crime prevention or as reaction of authorities to crime Vandalism of lighting
Commerce	causes	Crime	+++ -	Increased motivation and opportunity Increased employment reduces need
Crime	causes	Commerce	+ - -	Money-laundering investment Customers react and stay away Cost of security staff and insurance

4.4.3 Discussion of the lighting, commerce and crime hypothesis

Qualitatively at least, the hypothesised processes described appear sufficiently general and robust to cope with real-world variations with little effect on the overall longer-term outcome. For instance, the population increase in the state of Victoria during recent years consisted mostly of people older than 40 years, a group that contributes little to the overall crime rate (DCPC 2002, p 14). Regardless, both the crime rate and outdoor ambient light levels continued to increase substantially in Victoria in these years.[48]

DCPC (2001, p 14) discussed the observed attractiveness of urban centres with extended shopping hours and entertainment facilities, but missed the point that bright lighting appears to play an important or even essential role in the existence of this attractiveness. The combined effects of pleasure, arousal and dominance influence behaviour in particular environments (Mehrabian and Russell 1974). Mehrabian (1976) believed that lighting was a chief factor in the impact of the environment on individuals, and put this in the context of an inverted ‘U’ curve like the Yerkes-Dodson curve for human performance as a function of arousal. This carries an implication that sufficiently high absolute or relative levels of light may become unpleasant and repellent, and degrade visual performance, as is already well known.

Summers and Hebert (2001) found that moderate extra lighting of specific merchandise displays in shops did reliably affect customer behaviour in ways that appeared good for business. The positive effect of light on trade has been known at least qualitatively for a long time, and, reasonably, the lighting industry has prospered in meeting this need. Philips (2002) recommended shop interior illuminances of 300 to 500 lux, with spotlighting for specific displays. In the case of a jewellery shop,

“[An] entrance illuminated with a relatively high lighting level will serve to attract customers from a distance”,

and for a shoe shop,

“Well-designed lighting creates shop-windows that sell. Potential customers will be attracted to the window and then, it is hoped, persuaded to enter the shop. The main obstacle to achieving this is the reflections in the window caused by bright daylight. The task of providing enough brightness in the window is made particularly difficult because of the mostly dark colours of the merchandise”.

Rarely is there any appropriate dimming of the window display lighting at night (not to be confused with switching the display lighting off for part of the dark hours). Lit shop windows tend to be overbright at night if facing directly on to a street or other outdoor area.

The business world is also quite open about the key role of lighting in the profitability of shopping malls. For example, Horner (2002) described how an existing shopping mall lit internally to about 40 lux (“a dingy look”) was given an unstated increase of illumination. In the next two years, there was a one-third increase in traffic flow, sales increased 38%, fewer elderly citizens had slip and fall accidents and insurance premiums were reduced, and the result was a 19 percent increase in profit. Horner claimed that relighting of a car park at another shopping mall had reduced vandalism and made the area safer.

Presuming that the trials were uncontrolled, descriptions like this raise an issue about current business development practices. Little credence would be given in scientific work to experiments lacking adequate controls and without competent statistical analysis of probabilities of the results arising by chance. The results of such poor scientific practices are unreliable, and funding for further work would be jeopardised. Sectors of the business world seem unaware of or disinterested in such constraints, despite project expenditures that are often vastly greater than would apply to individual scientific projects. In this respect, practices in some business sectors need to catch up with the twentieth century if progress of these areas in the twenty-first century is to be based on more than the kind of unreliable assessments that really belong back in the nineteenth century or earlier. In particular, the business world will need to look systematically, through consultants where necessary, at ways of competing that do not unwittingly encourage the growth of crime and its attendant public and private costs.

There is a further observed effect of bright lighting installations in urban centres, viz a propensity for stepwise increases in the areas covered by lighting installations of given illumination levels. A suggested explanation is that residents and business proprietors located at and near the periphery of an urban centre might be supposed as believing themselves deprived of the ‘protection’ and customer-attracting aspects[49] of the brighter downtown lighting, and press for more public lighting in their area. This would usually come in due course because there are investments, jobs and especially votes at stake. It is more than just ‘keeping up with the neighbours’. Consequently, the suburbs tend to share in or follow the growth of city centre lighting through the physical spread of lighting installations[50] and successive upgrades in combination with the outwards optical spread of light. By processes like those governing the downtown crime rate, suburban crime rates tend to follow the urban crime rate as a more or less stable fraction of it, hardly a desirable result. At least at present, it is simply not practicable to continue the process outwards indefinitely by lighting rural areas to similar illuminance levels, and crime rates there remain lower again both for property crime and violent crime. This is as observed with a large data sample (Maguire and Pastore 2002, Table 3.121).[51]

At least in the longer term and for light levels at least those required for mobility safety, the net indirect all-hours parts of the lighting, commerce and crime interaction generally appear to override any direct net effects that lighting may have in hindering and facilitating crime at night. Therefore, controlled before-after studies like those used by Painter and others now appear to be of little value unless rigorous time-series spatial tracking of ambient light, commerce and crime is incorporated at the outset.[52]

Fortunately, quasi-experimental testing of the new hypothesis does seem possible with existing methodologies and manageable sample sizes and durations.[53] The implications of the hypothesis for urban growth control, decentralisation planning and urban design appear far-reaching indeed. Studies of the interactions in these areas may also assist understanding of lighting and crime interactions.

Economists and others have already found numerous examples of real-world variables with a coupled growth and two-way or indeterminate causal direction. A criminological example is the complex two-way positive and negative causal interactions of informal surveillance with each of robbery and burglary (Bellair 2000). Bellair also reviewed existing knowledge of a two-way causality between crime and fear of crime and interactions between these factors and others such as social networks and surveillance.

Another example is the link between police numbers and crime in the longer term (Marvell and Moody 1996, p 618). As it has been possible to get around the problem of apparent simultaneity in studying that relationship, it should be feasible to do the same with lighting and crime. For example, perturbing the relationship by the introduction of effective controls on lighting waste and other measures to reduce ambient outdoor lighting at night should allow quantification of any time delays and resultant effects on crime. Causal direction tests, such as the Granger test (eg Marvell and Moody 1996, p 617), may be applicable, subject to the caveats raised in the preceding section.

Historically, lighting has been increased as a crime prevention measure that now appears to be futile at best in its direct effect, while its indirect and counterproductive effects now appear to dominate, albeit largely via commerce. When other measures for controlling crime fall short of fully compensating for the indirect effect, further lighting increases continue the cycle, subtly making the crime problem worse. Urban and, increasingly, suburban commercial centres lead the way for suburbs and rural settlements with ever-brighter lighting. Politicians, urban designers, architects, the lighting and power industries, the advertising industry, business organisations and crime prevention practitioners all foster the trend to more and brighter lighting which results in more ambient light, more light trespass, more upward waste light, and ultimately, it now appears, more crime. The time is well overdue for such cycles to be stopped and their antisocial legacy effects not only to be contained but reversed.

What would be helpful now would be studies in which lighting increases were followed up not only for checking the stability of any changes in crime, but also to track developmental changes linked to the stimulus of brighter lighting. Painter and Farrington (2001b, p 9) explained why there were no follow-ups in Dudley and Stoke-on-Trent:

“It would have been desirable to investigate the permanence of the reductions in crime by conducting follow-up surveys, but unfortunately further environmental improvements were carried out in all areas which made it difficult to disentangle the effects of the improved street lighting.”

Further detail was given about the Dudley estate in Painter and Farrington (2001a, p 284):

“When the experimental estate was revisited in December 1994, it had changed out of all recognition (Painter 1995: 314). The Tenants’ Association, in conjunction with the Housing Department, had obtained £10m from the Department of the Environment for a programme of neighbourhood improvements. According to the Tenants’ Association, improved street lighting was the catalyst that signalled that the estate could be improved and that encouraged them to bid for more money. The fact that, at the time of the after survey, the estate was improving and was expected to improve even more probably led to increased community confidence and increased optimism by young people, including optimism by young people about finding a job. The changes in the experimental estate unfortunately made it impossible to carry out a follow-up study to investigate how far the effects of improved street lighting on crime persisted over time.”

Now, follow-up studies appear to be greatly desirable instead of “impossible”. Attention would need to be extended to the economics aspects of the changes that have taken place, insofar as these may have affected the incidence and type of crime.

Johnson, Bowers and Hirschfield (1997) studied repeated burglaries in the Merseyside area of the UK. Elevated risk of such repeat victimisation has been widely reported by others. The risk of a repetition reduces exponentially with time since the previous burglary. Johnson et al. offered an explanation that the victims would tend to install security equipment after a burglary and that as this might take time, it would favour early repeating. This implies that on repeat occasions, burglars target properties that are more likely to have security equipment than previously and more likely than is typical of properties in the area. Burglars could be expected to know this, which appears to run counter to the disproportionately high number of repeats. But if the addition of outdoor security lighting or increased street lighting is frequently among the reactive measures, the interpretation now is that the lighting increases the attractiveness of the property to the burglar. The matter is obviously complex, but this aspect would seem worth checking. Another aspect of the Johnson et al. paper is their use of the concept of burglary hotspots. It could be worth seeing if these areas are also a diffuse form of lighting hotspots.

Field (1999) developed a model of property crime linked to periods of economic growth and recession. Thefts and burglaries are linked to the stock of crime opportunities. In turn, these rise following a succession of good years. The rates for these crimes also rise with increases in the number of males aged between 15 and 20. On casual inspection, the similarity of graphs of the economic model and recorded thefts and robberies is compelling. Lighting is not in the model. However, expenditure on lighting could be expected to vary with economic conditions in a way that would tend to couple increased lighting with increased theft and burglary. Any causal relationships between lighting and acquisitive crime could tend to be concealed by co-varying factors in this type of analysis.

5. THE HYPOTHESIS AND FURTHER EVIDENCE

5.1 Lighting Constraints and Crime in San Diego

San Diego, CA, has had low-pressure sodium (LPS) as a large proportion of its outdoor lighting[54] for many years to try to minimise the adverse effect of its outdoor lighting waste[55] on the performance of the 5-metre (200-inch) Hale telescope on Palomar Mountain. Palomar and the Mount Laguna Observatory as well are both within 100 km of San Diego. Stray light from San Diego is the major source of artificial skyglow affecting the Hale telescope (Garstang 1989a). San Diego has been one of only a handful of places in California where light pollution, light trespass and glare have been addressed reasonably in the municipal code and in the county zoning ordinance, although over sixty other places in California are thinking about it (Skykeepers 2003).

San Diego is about 130 km from Los Angeles, which has three times the population. Using FBI (1998) Uniform Crime Reports (UCR) data, San Diego has a crime rate that is lower than for Los Angeles (11% lower for 4.51% against 5.07% for the respective cities, or 9% lower for 3.95% against 4.33% for the respective Metropolitan Statistical Areas). The difference is much larger when account is taken of the threat posed by the various crimes, as in the Morgan Quitno crime scores mentioned in Section 3.2.6 above.

San Diego is ranked 123 for crime safety on the Morgan Quitno (2000) list of 315 US cities with populations above 75 000. Los Angeles is ranked 222 (ie, less safe) on this same list. Differences in the Morgan Quitno scores and rankings are also in the direction expected on the basis of the new hypothesis, given that San Diego has had outdoor lighting constraints that have not been applied generally in Los Angeles.

San Diego recently decided to replace most of its 50-W low-pressure sodium streetlights by brighter, less well-shielded and more energy-consuming 150-W high-pressure sodium lights, supposedly to make the place safer for pedestrians at night (WalkSanDiego 2002) as part of the city’s growth strategy. This change was applied to within 50 km (30 miles) of Palomar despite strenuous objections by environmentalists and professional and amateur astronomers (eg Johnson 2001). More recently, the change was extended to within 25 km (15 miles) of Palomar (Monteagudo, 2003). The change is likely to be good for business, considered in isolation. From Part 1, the change is also likely to be futile at best for direct crime deterrence at night and environmentally damaging. According to the new hypothesis, as San Diego’s lighting characteristics move towards parity with those of the more conventional lighting in Los Angeles, San Diego’s crime rate could be expected to tend towards that of Los Angeles, hardly the result intended by the ‘more and brighter’ lighting proponents in San Diego. If the relighting decision is allowed to stand, collection of serial photometric and crime data well before and after the relighting should be considered, bearing in mind the problems of inadequately planned lighting and crime experiments described in Part 1.

Although the study of relatively isolated large-scale opportunities such as the San Diego relighting may lead to progress in knowledge of lighting and crime, any such advance may take several more years for sufficient time-series data to be collected. This presumes that adequate before data are available. As the plan is to replace the low-pressure lights over a five-year period, the absence of an abrupt change could reduce the sensitivity of the experiment and make generalisation of the results more difficult. Meanwhile, existing larger-scale evidence relevant to the hypothesis has been identified, and is now presented.

5.2 Crime and Measures of City Light

5.2.1 Satellite measures of upward light energy losses at night

Satellite measurements of artificial light radiated upward at night from populated areas of Earth are mentioned in Section 2.1.2 above. After the collapse of the Soviet Union in 1993, the US Air Force allowed the Operational Linescan System (OLS) of the US Defense Meteorological Satellite Program (DMSP) to be used on suitable occasions for non-military scientific observations. Although the nighttime part of the system was designed to detect clouds illuminated by moonlight, it was found that its sensitivity could be reduced sufficiently by ground control to allow mostly unsaturated detection of city lights on nights close to New Moon (eg Elvidge, Baugh, Kihn, Kroehl and Davis 1997; Cinzano, Falchi and Elvidge 2001). The resolution achievable at the ground is dependent on the viewing geometry from orbit. In the usual configuration, the projected (binned) pixel size is typically in the order of 2.8 km square, small enough to give useful indications of the energy emitted by aggregated artificial light associated with terrestrial human activity at various places. The threshold sensitivity corresponds to a population density of only about 8 persons per square kilometre (Nakayama and Elvidge 1999). The availability of data about the upward light emissions of cities and towns provides an opportunity to test the new hypothesis in the cases where crime data are also available. Before such tests are described, however, it is important to understand the limitations of the satellite data.

Isobe and Hamamura (1998) and Isobe (2000) listed upward light energy loss data for 153 cities from 52 different countries. They did not state whether these data were for:

· the DMSP OLS sensor response weighted by the luminosity function or similar to give the CIE luminosity response,

· the astronomers’ V (visible) band response, or, as appears more likely,

· left unweighted as a measure of radiant energy detected by the sensor, a multiplier phototube.

The sensor response extends from 0.47 µm to 0.95 µm (Chor-pang Lo 2002), which is not the 0.38 µm to 0.77 µm range of visible light. The sensor response is described as VNIR (visible and near infrared) by Elvidge et al. (1997). To add to the uncertainty, the data given by Isobe and Hamamura for cities in Turkey and elsewhere are repeated in a different paper (Aslan and Isobe 2003) with the same observational date, but for the Turkish cities only, the total energy losses are about 14 % larger and the losses per unit area are up to 43 % larger. Data attributed to Isobe and Hamamura are given by RASC (1999) for three additional cities in western North America, but data for other cities in the region are as much as 25% less or nearly double the values given for the same date in Isobe and Hamamura (1998) and Isobe (2000). The additional cities are not included in the following analysis.

The electrical industry units used for light energy by Isobe and Hamamura are customary rather than fully SI metric. They are reproduced here unchanged in two forms: one is electrical energy radiated above the horizontal at night per year in gigawatt–hours (the total light energy loss) for the whole city, and the other is this value divided by the city area in square kilometres. A third form, derived in this paper, is the total light energy loss divided by the city population to give per capita values. If these quantities did truly represent energies of visible light for given operating durations, the first and third would be proportional to luminous intensity and the second, with a different factor of proportionality, to mean luminance.[56] The quantities involved are treated here as though they are photometric quantities, although the approximation is crude.

The approximation is even poorer than is apparent because the photopic (cone vision) spectral response implied is rather inappropriate. Given the generally low light levels involved, visual quantities might more usefully be described in scotopic (rod vision) terms, or at least in mesopic terms, a transitional mix of photopic and scotopic characteristics. By comparison with the photopic spectral response, the scotopic response is displaced towards the violet end of the spectrum. In this dim-light adaptive state, not only does the eye remain insensitive to near infrared but it is also insensitive to red light. One consequence of this is that cities that radiate relatively more incandescent lamp light and less gas-discharge lamp light than cities with a typical mix would be visually fainter than would be indicated by the satellite measures. Different kinds of lamps are in common use for outdoor purposes and their distinctive spectra have been observed, separately and mixed, in airborne calibration of DMSP OLS nighttime data (Elvidge and Jansen 1999). The present analysis does not take between-city differences of lamp-type mixes into account as the information does not appear to be readily available, if at all.

For brevity, the measures derived from satellite-based optical radiation are called light energy losses in this paper. Given its exploratory nature, the analysis presented is considered sufficiently robust for its purpose. In due course, upwardly radiated light losses will be presumably become available as true photometric quantities. This could be expected to reduce confounding contributions to the variance in correlations between light and crime quantities.

The publication of satellite measures of upward light energy losses has stimulated debate on energy conservation and greenhouse gas emissions relating to outdoor lighting. Unfortunately, many of the arguments put to date have not taken account of inefficiencies in the conversion of electrical energy into light energy. A perfectly efficient electric lamp would convert 1 watt of electrical energy into 683 lumens of green light (555 nm), or a lesser number of lumens of white light, the amount depending on the spectral energy distribution. Actual lamps tend to be much less efficaceous than this in producing their visible light output. The actual electrical energy inputs required to produce the observed upward light losses are expected to be between about five and ten times greater than the energy values given by Isobe and Hamamura.

5.2.2 Upward light energy losses for various cities

The upwardly radiated light from a city, town or other populated area includes a substantial component of unused waste light that is emitted above the horizontal directly from outdoor light sources. These sources include advertising signs, road signs, traffic signal lights, and vehicle lights. It is convenient to include here also the light radiated above the horizontal from external windows of internally illuminated buildings, although part of it is used waste, ie it has performed a useful function indoors. The balance consists of waste used light that has been reflected above the horizontal from the terrain and built environment. As a first approximation, the total light in space detected coming from a populated area is proportional to the total amount of outdoor lighting within the area. Obviously, factors such as presence and effectiveness of luminaire shielding,[57] extent of use and opacity of drapes and blinds at windows, and the mix of light-coloured concrete and blacktop road and path surfaces will affect the constant of proportionality. For the immediate purpose, accurate relative measures of upward light energy loss between cities appear to be a sufficient guide to the relative amounts of ambient artificial light available for human outdoor activities at night.

The Isobe and Hamamura data appear to have an acceptable degree of internal consistency, although the authors warn of errors from sensor saturation by the bright centres of some cities. A related point is that the nominal surface resolution of the satellite OLS system, 2.8 km, is somewhat too large for accurate measures of the bright central lighting peaks of many cities (eg NASA 2000).

These shortcomings in the data could result in local energy loss underestimation. Of the 153 cities listed, the ten cities with the largest amounts of upward light energy loss per square kilometre are shown in Table 3, along with the ten having the smallest amounts. The observed range in values is remarkable, even allowing for the effect of snow cover in inflating some of the values (discussed below). Note that a high or low ranking for light loss per unit area is not necessarily a good predictor of light loss per person.

TABLE 3. World Cities: Upward Light Energy Losses and Population
City, Country	Population, thousands (Year)	Annual Upward Light Energy Loss per Unit Area, MW.h/km²	Annual Upward Light Energy Loss per Person, kW.h
Ten highest light energy loss/km²:
Trois Rivières, Canada	138 (2001)	205	53.4
Clermont-Ferrand, France	1 310 (1999)	87.0	4.45
Ciudad Juarez, Mexico	1 187 (2000)	74.2	13.1
Calgary, Canada	951 (2001)	43.8	87.7
Montreal, Canada	3 426 (2001)	34.4	40.6
Edmonton, Canada	938 (2001)	32.4	62.8
Toronto, Canada	4 683 (2001)	31.6	29.3
Minneapolis MN, USA	368 (1999)	28.2	332
Las Vegas NV, USA	715 (1999)	24.5	53.1
St Louis MO, USA	397 (1999)	22.9	234
Ten lowest light energy loss/km²:
Konya, Turkey	743 (1999)	2.11	1.57
Kochi, Japan	331 (2000)	1.96	4.32
Gold Coast, Australia	404 (2001)	1.77 (1996)	4.26
Wellington, New Zealand.	167 (2001)	1.67 (1996)	7.19
Belfast, Northern Ireland	279 (1999)	1.52	4.52
Antalya, Turkey	603 (1999)	1.41	1.63
Christchurch, New Zealand.	324 (2001)	1.31 (1996)	5.19
Brasilia, Brazil	1 960 (2000)	1.19	3.27
Phnom Penh, Cambodia	920 (1994)	0.92	0.55
Pyongyang, North Korea	2 360 (1987)	0.11	0.0061
The light energy losses are based on measurements of 153 cities by satellite, in early 1997 unless otherwise indicated (Isobe and Hamamura 1998). Corrections to no-snow conditions are not made in this table. Populations for the years indicated are either from national statistical offices, Brinkoff (2002) or van der Heyden (2002).

A problem with the Isobe and Hamamura data now becomes evident in relation to one of the cities in Table 3. Las Vegas, Nevada, is well known for its numerous intensely bright outdoor advertising lights, signs and laser displays (ILDA 2002). The scattered light dome above the city is growing rapidly and beginning to affect the formerly pristine night sky conditions over Death Valley National Park (Albers and Durisco 2002). “As seen from space, Las Vegas is the brightest city on earth [sic]” (Schweitzer and Schumann 1997). But Table 3 indicates differently: on a light per unit area basis, Trois Rivières in Canada is 8.4 times brighter than Las Vegas, which ranks only ninth in the Isobe and Hamamura list. Apart from the sensor saturation problem, parts of these discrepancies are doubtless due to difficulties of matching areas and populations between space and ground measures. Another possible reason is that the cities brighter than Las Vegas might have had snow cover, which would make these cities appear brighter to the satellites. Las Vegas itself did not have snow cover.

Appropriate online or hard copy historical weather data could not be found for Clermont-Ferrand. This city is on high ground, however, and the satellite observation was made on 1997-01-13, ie during the northern hemisphere winter. Ciudad Juarez has snow only on about 2 days a year, but has 53 icy days/year (Ciudad Juarez 2001). The satellite observation was on 1997-01-11. Wunderground (2002) gave details for the day: the minimum temperature was 4.0°C, no snow event was recorded and no snow depth was given. It is possible that some frost was on the ground.

Wunderground (2002) also gave details for the day of observation for the Canadian cities in Table 3, except Trois Rivières. They had thick snow cover recorded with the exception of Montreal, which did have a snow event on the day, however. Minneapolis also had thick snow cover.

An attempt is made here to estimate what effect the presence of snow might have had on the satellite measures. Newly fallen clean snow has a visible total reflectance of over 90% (eg Leeds 2003). Subsequent contamination of the surface, eg by dust and soot, can reduce this to as low as 20% (Klein, Hall and Nolin 2000) or even lower if the snow turns to dirty ice. The reflectance of the typical built environment is about 10%. At first sight it might be thought that snow cover could increase the satellite measure by up to nine times but this is without taking account of important factors such as reduced reflectance of snow in the near infrared (ie within the DMSP OLS sensor spectral response region), and the waste light radiated above the horizontal by illuminated signs, internally lit windows and fixed inadequately shielded or poorly aimed luminaires. Snow may also discourage or prevent the use of motor vehicles and thereby reduce the total amount of light radiated by vehicle lights.

Assume that the upward waste light ratio (UWLR) for streetlights and other public lighting is 0.15. (Typically it is zero for full-cutoff types,[58] less than 0.15 with semi-cutoff types and more for mercury vapour ‘flower pots’, globes and wall packs.) Assume that the UWLR for all advertising signs, floodlights, decorative lighting and external windows is 0.5, and for all vehicle lights, 0.1. Denote the total source flux in lumens as S for street and path lights, A for signs etc. and V for vehicle lights. Then the total direct upward flux T_DU from these sources is

T_DU = 0.15 S + 0.5 A + 0.1 V.

The total upward flux reflected from the built environment T_RU is the total downwards flux multiplied by the effective reflectance of the terrain, R_T:

T_RU = (0.85 S + 0.5 A + 0.9 V) R_T.

The total upward flux is

T_U = T_DU + T_RU.

As an estimate of typical conditions, put A = 0.3 S and V= 0.1 S. Then the total upward flux becomes

T_U = (0.31 + 1.09 R_T ) S.

But the total flux emitted by all sources, T, is given by

T = S + A + V, ie 1.4 S.

Therefore the fraction of total flux that is directed upward, the Upward Fraction UF is

UF = T_U / T = 0.221 + 0.779 R_T.

As a check, UF = 1 when R_T = 1. This indicates, correctly, that with a perfectly reflecting terrain, all emitted light would eventually travel in directions above the horizontal.

A typical value for R_T in a city is about 0.1. This results in 29.9% of the total light being radiated above the horizontal, consistent with the fraction (1/3) generally thought typical with present inefficient lighting practices. Of the Upward Fraction in this example, 0.221/0.299 or 74% consists of light directly radiated above the horizontal from the light sources, ie unused light waste. The remainder is used light waste. This is consistent with the impression that the light seen in close city views from aircraft at night mostly comes directly from luminaires, unshielded lamps, undraped windows and illuminated signs rather than from illuminated areas such as paved surfaces, walls and vegetation.

If all such upward unused waste light were absorbed by a hypothetical instant installation of full-cutoff shields (somewhat impracticably including all advertising signs, floodlit structures etc.), then skyglow would be reduced immediately to 26% of its former value, all else remaining unchanged. Given that skyglow is typically increasing by about 10% or more a year, it would take only about 14 years or less for the skyglow to reattain its previous value. The exponential growth would then resume its increase beyond the level it was at when interrupted by the full-cutoff transformation. This is why a permanent solution to the skyglow problem must involve mandatory caps on total outdoor light flux or energy use as well as restrictions on direct light emission above the horizontal.

If the instant shielding introduction were restricted to streetlights, the immediate reduction in skyglow would be to 0.107/0.299 or to 36%. This would give less than 5 years of respite from the growth of skyglow. The time would be even shorter if the mean terrain reflectance were higher than 0.1, eg 4 years if R_T = 0.15.

The Upward Fraction is shown in the following table for various values of R_T, along with the ratio of UF to its value for R_T = 0.1. This ratio shows how increased values of R_T would increase the upward flux measured by a satellite.

TABLE 4. Upward Light Loss as a Function of Terrain Reflectance
Terrain Reflectance R_T	0.05	0.1	0.2	0.3	0.6	0.7	0.8	0.9
Upward Fraction UF	0.26	0.299	0.377	0.455	0.688	0.766	0.844	0.922
UF / UF_0.1	0.86	1	1.26	1.52	2.30	2.56	2.82	3.08

The values of the Upward Fraction are fairly insensitive to the assumptions made, within ranges that might be expected in practice.

A fresh snowfall covering all upwardly facing surfaces and viewed from directly overhead could be expected to have R_T = 0.9 approximately, thereby increasing the apparent light flux from a city by about 3.08 times. It would be less than this if seen from appreciable zenith distances because of the area of building walls visible and not covered by snow. Other factors could also reduce the effective average terrain reflectance, such as cleared roads and paths, and dirty, thin, patchy, consolidating or melting snow. All of these effects can be taken account of by using an appropriately lower value for the reflectance.

Accordingly, the net effect of complete fresh snow cover is estimated as an increase of about three times in upward light energy loss, although a factor of four has been used by others in discussions of city lights and satellite measurements. Using a factor of three or even four still leaves substantial discrepancies between the claim about Las Vegas and the data for the first three cities in Table 3.

It might be thought that suburban lights seen from space would appear to extend further into the countryside when snow cover increases the Upward Fraction. This would be true if the light sources on the ground gradually became more spread out with distance from the city centre. However, lit features typically tend to be sharply delimited according to Elvidge et al. (1997). The thresholding process for city light limits is described in NASA (2000). Planning controls on land use would appear to be a factor in the relatively well defined edges observed. Natural boundaries such as waterways and steep changes in terrain level could also contribute. The point here is that overestimation of city area does not seem to be an important effect resulting from snow cover and therefore does not help to explain why Las Vegas is not at the top of the list in Table 3.

If there are problems with the internal consistency of the satellite data, the effect would be to increase the unexplained variance in lighting and crime correlations such as those described below. For the present, the anomaly has to be left unresolved. This does not prevent further use being made of the data.

Unfortunately, crime data for many of the 153 cities listed by Isobe and Hamamura are either incomplete or apparently not available or not readily comparable with those of cities in other countries because of differing criminal justice systems (eg the Scottish system compared with that of England and Wales, or the New Zealand system in comparison with Australia’s). Both crime and light loss data are required for several cities from any particular country as a necessary but not sufficient condition to give a sufficiently homogeneous data set for analysis. The quantity of data pairs in the set is primarily limited by the number of cities with light energy loss data and the quality mainly relates to the national consistency of crime data. This condition appeared to be best met by the USA city data set. Data sets for Canada and the UK were about equal in quantity and apparent quality, with the data set for Australia not as good mainly because of crime data quality issues. It would seem reasonable to expect that there are other usable city data sets yet to be identified.[59] Lack of concurrency of the light and crime data may be more of an issue with the data sets from Canada, UK and Australia.

The order of presentation of the light energy loss and crime results in the following sections is USA, Canada, UK and Australia.

5.2.3 City crime and upward light energy loss comparisons, USA

5.2.3.1 UCR crime rate data and light energy loss per unit area

Table 5 lists crime rates and annual upward light energy loss per square kilometre of city area for all 21 of the USA cities included in Isobe and Hamamura (1998). In most cases, the Uniform Crime Reports (UCR) tables provide populations and crime numbers for cities, along with populations, crime numbers and crime rates for larger areas that include the cities. The populations and crime rates in the table are values for, or calculated from, the city entries alone.

An expectation from the new hypothesis is that, all else being equal, cities with large upward light energy losses per unit area or per person will tend to have higher crime rates than cities with smaller losses. An initial test of the USA data is made by plotting the UCR Index crime rate for each city (FBI 1998) against annual light energy loss per square kilometre (Figure 7).[60] The linear regression line slope of 0.031 is in the expected direction, but r² is negligible at 0.0094. Two-tail tests for statistical significance are used throughout this Chapter 5. In this case, the slope is not significantly different from the ‘no correlation’ slope of 0 (t = 0.426, 19 df).

TABLE 5. Crime and Upward Light Energy Loss in USA Cities
City	UCR Index Crime Rate %	Morgan Quitno Crime Rate Score	Morgan Quitno Crime Rank: Safe 1, Unsafe 315	Annual Upward Light Energy Loss per Unit Area MW.h/km²
Year	1998	1998	1998	1997
Minneapolis MN	9.561	183	293	28.2 (11.0)
Las Vegas NV	5.846	78.34	225	24.5
St Louis MO	13.53	302.3	311	22.9 (11.2)
Denver CO	5.306	51.74	195	18.4 (12.1)
Philadelphia PA	7.319	163.6	287	18.0
Buffalo NY	7.164	110.9	257	16.0 (10.5)
Baltimore MD	10.94	307.6	312	15.7
Kansas City MO	11.79	236.9	304	15.5 (10.2)
New York City NY	4.392	51.62	194	15.0
Washington, DC	8.828	230.9	299	13.5
Boston MA	6.251	81.67	231	13.4
Phoenix AZ	8.545	89.28	241	11.5
Boulder CO	5.200	-39.71	46	8.38
Sacramento CA	8.219	91.08	243	7.72
Stockton CA	7.311	81.32	230	7.38
Tucson AZ	9.685	96.32	249	7.32
Modesto CA	6.737	18.67	142	6.77
Fresno CA	7.933	85.41	235	6.67
Portland OR	9.548	99.36	251	4.51
Salem OR	8.375	8.69	126	4.43
Eugene OR	9.013	13.74	136	2.73
The entries are for all 21 USA cities included in Isobe and Hamamura (1998). They are in descending order of annual upward light energy loss per unit area, as observed by satellite. Estimated values for no-snow conditions are shown in parentheses. Population and UCR crime rate data are from FBI (1998) except for St Louis, Buffalo, and Kansas City, which were missing; instead, their UCR data for 1998 are from FightCrimeFortWayne (1999). The Morgan Quitno (2000) score of weighted UCR crime rate is positive if above and negative if below the mean weighted UCR crime rate. The crime ranking is based on this score for 315 US cities. The UCR crime rate is plotted against upward light energy loss per unit area in Figure 7. The crime score is used in later figures.

TABLE 6. Population, Crime and Upward Light Energy Loss in USA Cities
City	Population, thousands	UCR Index Crime Number, thousands	Annual Upward Light Energy Loss, GW.h	Annual Upward Light Energy Loss per Person, kW.h
Year	1998	1998	1997	1997/1998
New York City NY	7 358	323.2	136	18.5
Minneapolis MN	362.1	34.62	122 (47.7)	337 (132)
St Louis MO	344.2	51.46	93.0 (45.6)	270 (132)
Kansas City MO	447.7	53.73	71.5 (47.0)	160 (105)
Phoenix AZ	1 226	104.7	55.0	44.9
Philadelphia PA	1 449	106.1	48.5	33.5
Washington DC	523.0	46.17	41.8	79.9
Las Vegas NV	908.6	53.12	38.0	41.8
Denver CO	509.3	27.03	29.4 (19.3)	57.7 (38.0)
Baltimore MD	662.3	72.50	29.2	44.1
Portland OR	488.8	46.07	22.2	45.4
Buffalo NY	308.6	22.32	20.0 (13.2)	64.8 (42.6)
Boston MA	559.6	34.98	15.0	26.8
Sacramento CA	384.7	31.62	14.9	38.7
Tucson AZ	467.7	45.30	13.2	28.2
Fresno CA	404.3	32.08	6.66	16.5
Stockton CA	239.7	17.53	3.68	15.4
Modesto CA	183.3	12.35	3.03	16.5
Salem OR	126.4	10.58	2.36	18.7
Eugene OR	127.2	11.46	2.09	16.4
Boulder CO	94.21	4.899	1.27	13.5
The cities in Table 5 are here listed in descending order of their annual total upward light energy loss observed by satellite (Isobe and Hamamura 1998). Estimated values for no-snow conditions are shown in parentheses. Population and number of UCR Index crimes are from the sources given in Table 5. Population is plotted in Figure 8 against light energy loss per unit area from Table 5. The no-snow light energy loss data are plotted in Figures 9 and 10.

5.2.3.2 Investigation of apparent outliers

Table 6 contains additional information relating to crime and light energy loss for USA cities.

The annual total upwardly radiated light energy loss values in Table 6 are analogous to luminous intensity maintained for a given time. To test for possible anomalies in the DMSP data, these values were plotted in Figure 8 against city population for each of the 21 cities. Many others have already made plots of light losses against demographic and related variables. NASA (2000) stated that the areal extent of city light sources is highly correlated with electric power consumption and population. The total light emitted by each city is less well correlated with population (Isobe and Hamamura 1998), as is indicated by the nearly six-hundredfold range of variation shown in the rightmost column of Table 3, without considering Pyongyang. This suggests that light energy loss per person could be worth testing for correlations with crime.

A plot similar to Figure 8, but for many more cities, is given by Cinzano (2000a). The regression line for the Cinzano plot passes close to the data point for New York City, which therefore appears to be reasonably representative of world cities in terms of total light loss per person. The upper right data point in Figure 8 is for New York City, so it marks one end of a trendline based on world data.

Any regression line for Figure 8 is constrained to pass through the origin, so the world trendline is indicated well enough by the diagonal for the figure. Many of the US cities are on the bright side of the trendline. But the three data points low down and in between 60 and 130 on the light loss scale are so bright that they can be regarded as outliers. From the right, the three cities are Minneapolis, St Louis and Kansas City.

A plot of crime numbers against total light energy loss (not shown) indicates that these three cities are again outliers, with uncharacteristically low numbers of crimes for a given amount of upward light energy. This could mean that lots of upward light, implying brightly lit cities, is associated with reduced crime. But if crime in these cities is not unusually low on some measure such as crime rate, it could mean that crime is not being affected much or at all by excessive lighting. Alternatively, it could mean that the overall amount of light has been increased well ahead of the numbers of crimes as a transient non-equilibrium stage, or that anti-crime actions such as intense policing and incarceration, unrelated to lighting, are having a beneficial effect in reducing crime.

Reference back to Figure 7 and Table 5 indicates that of the 21 cities, Minneapolis, St Louis and Kansas City are ranked 1, 3 and 8 respectively in terms of upward light energy loss per unit area, and 5, 1 and 2 in terms of UCR Index crime rate. The light energy loss ranking is 2, 3 and 4 on the total light scale. The actual crime rates are therefore relatively high, and this is evidence against the notion that these cities have found a way of using lots of light to reduce crime. Reasons for the large upward light energy losses remain unknown, but lots of upwardly aimed floodlighting could be suspected as a contributing cause.

5.2.3.3 Light-loss correction for snow cover

US city weather records (Wunderground 2002) were checked for snow cover on the dates given for satellite data collection. Minneapolis had 410 mm, St Louis 150 mm, and Buffalo 30 mm. Although there were no data available for Central Park in New York City, JFK airport recorded no snow cover. Rain/snow events were recorded for Baltimore, Boston and Washington DC, but no details were given for snow cover. Philadelphia had rain only. No snow-cover details were given for Kansas City but temperatures had remained well below freezing for the whole of the time since snowfalls three days earlier. Records from NOAA (2002) showed that up to 75 mm of snow/ice was present on the ground at times at one Kansas City weather station on 1997-01-11, -12 and -13, and traces to smaller amounts at two other stations. Climate maps indicated that other cities in Tables 5 and 6 were less subject to snow cover. Nevertheless, all were checked against the Wunderground historical data. No details were available for Boulder. Denver had 30 mm snow depth on the satellite measurement day. The remaining cities had no snow events and no snow cover or no record of any snow cover.

On the assumption that roads and paths had been cleared of snow and that the snow was contaminated by vehicle exhaust particles, the effective terrain reflectance was estimated to be 0.7 for Minneapolis and 0.6 for St Louis. The light energy loss values were divided by 2.56 and 2.04 respectively (from Table 4) to give values more likely to be representative of no-snow conditions. In the case of Buffalo, Denver and Kansas City, the effective terrain reflectance was thought more likely to be about 0.3 at the time of measurement. The factor for light energy loss correction to no-snow conditions for these three cities was therefore 1.52. The corrected data points were closer to the notional trend line of Figure 8, but still to its right (brighter) side. The corrected light energy losses for Minneapolis, St Louis, Buffalo, Denver and Kansas City are shown in parentheses in Tables 5 and 6.

The linear regression analysis was repeated for the plot values of Figure 7 after the corrections to give no-snow conditions were applied to the five cities. The slope became negative and larger, -0.102, but was still not significantly different from zero (r² = 0.052, t = 1.016, 19 df).

Neither Figure 7 nor its version corrected to no-snow conditions (not shown) represents a year’s results accurately, as snow was present only for part of the year. Presumably they do represent their respective parts of the year well enough, and the whole year’s results would be an appropriate mix of the two parts, weighted according to the fraction of days with snow cover. Clearly, the regression line slope for a combined result would still not be significantly different from zero.

The sample size for Figure 7 is the maximum set by the available source of light energy loss data. The choice of crime data is less constrained, and may have a large effect on the results.[61] Although the 1997 UCR report is still available (FBI 1997), it was not used as it has no data for Buffalo, Kansas City, Las Vegas, Philadelphia and St Louis. These five cities were in the eight brightest of the 21 and their absence would have biased and degraded the results. Given the nature of the hypothesis, it seems quite reasonable to use crime data for a year (1998) that started ten or eleven months after the satellite measurements of light energy loss.

5.2.3.4 UCR crime rate data and light energy loss per person

Figure 9 is also a plot of the UCR crime rate, but this time the abscissa represents upward light energy loss per person (from Table 6), corrected as described to no-snow conditions. The regression line is positive and its slope is reliably different from zero (r² = 0.459, t = 2.953, 19 df, p < 0.01). Without the snow correction, the regression line slope is smaller but still positive and statistically significant (r² = 0.338, t = 2.708, 19 df, p < 0.05). The changes introduced by the corrections to no-snow conditions are relatively insensitive to the assumptions made in deriving the corrections. As before, the actual results for the whole year would be intermediate between the original and no-snow values.

Of the 21 USA cities, those with higher values of UCR crime rate have reliably more upward light energy loss per person. This supports the existence of coupled growth of lighting and crime and the lighting, commerce and crime hypothesis devised to explain it, but, by itself, does not allow assignment of causality. Regardless, the regression equation shown on Figure 9 appears to have a useful amount of predictive power in determining the effect of a US city lighting change on the UCR index crime rate. Within the data set examined, the equation accounts for 46% of the variance in the crime data.

The data in Tables 5 and 6 were used to make a graph (not shown here) of UCR index crime rate against population. Logarithmic transformation was used to deal with the gap between the population of New York City and other cities. Despite the common belief that bigger cities have more crime, the slope in this case was slightly negative. No reliable connection exists between UCR crime and log₁₀ population for the 21 cities: r² = 0.039 and t = 0.876, 19 df, ns. This helps in interpretation of the findings for light and crime.

5.2.3.5 Morgan Quitno crime data plots

As mentioned in Section 3.2.6, the Morgan Quitno scores are based on UCR crime rate data weighted by the posed threat for various crimes, determined by survey. Cities and metropolitan areas in the USA with populations of more than 75 000 are included. A score is expressed as a positive real number when the weighted UCR value is greater than the mean weighted crime rate for all of the cities. A score of 0 means the place is representative of the mean, while a negative value means less dangerous than the mean. The Morgan Quitno crime scores for 1997 were available but were missing those cities mentioned above as missing from the 1997 UCR data. The Morgan Quitno (2000) crime scores for 1998 and their ranking are used instead in this paper, and are both listed in Table 5 for the 21 cities.

The Morgan Quitno crime scores were plotted (not shown here) against light energy loss per square kilometre. The regression line had a reliably positive slope (7.225) using the uncorrected light loss data (r² = 0.292, t = 2.356, 19 df, p < 0.05) and a statistically non-significant lesser positive slope (6.585) when the light measures were corrected as described to no-snow conditions (r² = 0.127, t = 1.551, 19 df). The statistical significance of the combined result would depend on the duration of snow cover over the year for each of the five cities observed with snow cover.

The Morgan Quitno crime scores were then plotted against light energy loss per person. The regression line had a reliably positive slope of 1.886 using the light loss data corrected to no-snow conditions (r² = 0.537, t = 3.194, 19 df, p < 0.01) (Figure 10) and a lesser but still reliably positive slope with data uncorrected for snow cover (slope = 0.662, r² = 0.368, t = 2.644, 19 df, p < 0.05) (not shown). A weighted combination representing a year would certainly retain statistical significance.

This finding is of far-reaching importance: USA cities with higher values of Morgan Quitno crime scores among the 21 have reliably more upward light energy loss per person. As with the finding for Figure 9, it supports the existence of coupled growth of lighting and crime.

Slightly smaller positive correlation coefficients result when the ordinates are rankings, ie ordinal data. This is not surprising given that the ranking is out of 315 and the ranks have a close monotonic relationship with the scores. An even higher correlation coefficient, r = 0.797, is obtained when the plot (not shown) is for Morgan Quitno scores against log₁₀ (no-snow light energy loss per person): r² = 0.635, t = 3.473, 19 df, p < 0.01. This indicates that the relationship between the crime score and light energy loss per person is non-linear. Another way of interpreting this result is that the crime score increased with the perceived amount of light per person.

The evidence for the lighting and crime connection appears to be even stronger than is indicated by Figure 10 and the preceding paragraph when the circumstances are considered in more detail. Firstly, most of the cities listed by Isobe and Hamamura (1998) appear to have been selected for measurement on the basis of being relatively bright and well defined in spatial extent. The Isobe and Hamamura maps of light energy loss distribution show many discrete areas fainter than most of those included in their listed results. Likewise, many cities and towns showing as patches of light are identified in a map of central England by Cinzano (2000a), and only the brightes

USAGE	TERM IN FULL [PLACE OR EXPLANATION]

AAA	American Automobile Association
ABS	Australian Bureau of Statistics
ACT	Australian Capital Territory
ad hoc	arranged for this purpose, specific
Amherst PD	Amherst Police Department [NY, USA]
AS	Australian Standard
AS/NZS	Australian and New Zealand Standard
AUD	Australian dollars
AVIRIS	Airborne Visible/Infrared Imaging Spectrometer [remote sensing]
AZ	Arizona [USA]
BAA	British Astronomical Association
BHP	Blackout History Project [USA]
BIS	British Information Service
BJS	Bureau of Justice Statistics [USA]
BOCSAR	Bureau of Crime Statistics and Research [NSW, Australia]
CA	California [USA]
CBD	central business district
ccd	charge-coupled device [image sensor in video and digital cameras]
CCP	Cities for Climate Protection [international organisation]
CCTV	closed circuit television
cd	candela, the SI metric unit of luminous intensity
CfDS	Campaign for Dark Skies [British Astronomical Association, UK]
CIE	Commission Internationale de l’Eclairage, or International Commission on Illumination
CO	Colorado [USA]
CPR	cross-product ratio
CPRE	Campaign to Protect Rural England
CPTED	Crime Prevention Through Environmental Design
CRCIT	Crime Reduction College Information Team [UK]
CT	Connecticut [USA]
DC	District of Columbia [USA]
DCPC	Drugs and Crime Prevention Committee [Victoria, Australia]
df	degrees of freedom [in statistics]
DMSP	Defense Meteorological Satellite Program [USA]
DOE	Department of Energy [USA]
EDT	Eastern Daylight Time [USA]
EST	Eastern Standard Time [USA]
eg	exempli gratia [for example]
et al.	et alii [and others]
F	Snedecor’s variance ratio
FBI	Federal Bureau of Investigation [USA]
FCO	full-cutoff [type of luminare]
FL	Florida [USA]
FS	fully shielded [type of luminare]
GW.h	gigawatt-hours [conventional unit of electrical energy]
h	hour
HPS	high-pressure sodium (lamp)
HSHF	Home Sweet Home Front [UK]
ICOLE	Indiana Council on Outdoor Lighting Education [USA]
IDA	International Dark-Sky Association
ie	id est [that is]
IESNA	Illuminating Engineering Society of North America
ILDA	International Laser Display Association
ILE	Institution of Lighting Engineers [UK]
IMHO	in my humble opinion
in toto	completely
JFK	John F. Kennedy [airport]
JRSA	Justice Research and Statistics Association [USA]
klm	kilolumen [1000 times the SI metric unit of luminous flux]
km	kilometre[1000 times the SI metric unit of distance]
KS	Kansas [USA]
kW	kilowatt [1000 times the SI metric unit of energy]
LA	Los Angeles [CA]
lm	lumen [the SI metric unit of luminous flux]
log, log₁₀	logarithm, logarithm to the base 10
log units	powers of ten change in a physical or psychophysical quantity
LPS	low-pressure sodium (lamp)
lux	SI metric unit of illuminance [1 lux = 1 lumen per square metre]
m	metre, or milli [one thousandth; wrongly, million]
m²	square metre
M	mega, or million
MA	Massachusetts [USA]
mcd/m²	millicandela per square metre [submultiple, SI unit of luminance]
MD	Maryland [USA]
MI	Michigan [USA]
MIT	Massachusetts Institute of Technology
MN	Minnesota [USA]
MO	Missouri [USA]
MODIS	Moderate Resolution Imaging Spectrometer [satellite instrument]
MQ	Morgan Quitno [company, USA}
MW.h/km²	megawatt-hours per square kilometre [energy per unit area]
NAPBC	National Action Plan on Breast Cancer [USA]
NASA	National Aeronautics and Space Administration [USA]
NBI	New Buildings Institute [California, USA]
NCJRS	National Criminal Justice Reference Service [USA]
NELPAG	New England Light Pollution Advisory Group [USA]
NHW	Neighbourhood Watch
NIJ	National Institute of Justice [USA]
nL	nanoLambert [submultiple of customary unit of luminance]
nm	nanometre [one-billionth (10^-9) of a metre]
ns	not significant
NSW	New South Wales [State, Australia]
NV	Nevada [USA]
NY	New York [State, USA]
NYC	New York City [USA]
NZ	New Zealand
OCS	Office of Crime Statistics [South Australia]
OESR	Office of Economic and Statistical Research [Qld, Australia]
OLS	Operational Linescan System [of the DMSP]
ONS	Office of National Statistics [UK]
OR	Oregon [USA]
p	probability
PA	Pennsylvania [USA]
Qld	Queensland [State, Australia]
r	ordinary least squares correlation coefficient
r²	square of the correlation coefficient
R_T	effective reflectance of the terrain
SA	South Australia [State] or Standards Australia
SCO	semi-cutoff [type of luminare]
SCP	Situational Crime Prevention
SI	Système Internationale or International System [of units]
t	Student’s t statistic
TPD	Tucson Police Department [Arizona, USA]
UCR	Uniform Crime Report(s/ing) [USA]
UF	Upward Fraction [of light]
UK	United Kingdom
ULR	Upward Light Ratio [of a luminaire]
UN	United Nations
UNESCAP	UN Economic and Social Commission for Asia and the Pacific
US, U.S., USA	United States of America
USGBC	U.S. Green Building Council
UT	Universal Time [formerly astronomical standard time]
UWLR	Upward Waste Light Ratio [of a luminaire]
VA	Virginia [USA]
Vic	Victoria [State, Australia]
Vicpol	Victoria Police [Australia]
viz	videlicet [namely]
VNIR	visible and near infrared [DMSP OLS sensor response range]
WA	Washington [State, USA]
WA	Western Australia [State, Australia]
WAPS	Western Australia Police Service
WI	Wisconsin [USA]
WW2	World War 2
yr	Year

µcd/m²	microcandela per square metre [submultiple, SI unit of luminance]
µm	micrometre [one-millionth (10^-6) of a metre]